CN115579108A - AI hemodialysis scheme model modeling method and method for generating hemodialysis scheme - Google Patents
AI hemodialysis scheme model modeling method and method for generating hemodialysis scheme Download PDFInfo
- Publication number
- CN115579108A CN115579108A CN202211418479.1A CN202211418479A CN115579108A CN 115579108 A CN115579108 A CN 115579108A CN 202211418479 A CN202211418479 A CN 202211418479A CN 115579108 A CN115579108 A CN 115579108A
- Authority
- CN
- China
- Prior art keywords
- component
- dialysate
- algorithm
- dialysis
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 96
- 230000000322 hemodialysis Effects 0.000 title claims abstract description 76
- 238000001631 haemodialysis Methods 0.000 title claims abstract description 74
- 238000000502 dialysis Methods 0.000 claims abstract description 482
- 239000003795 chemical substances by application Substances 0.000 claims abstract description 33
- 239000003814 drug Substances 0.000 claims abstract description 16
- 238000004422 calculation algorithm Methods 0.000 claims description 518
- 238000004458 analytical method Methods 0.000 claims description 231
- 239000011575 calcium Substances 0.000 claims description 120
- 239000011734 sodium Substances 0.000 claims description 118
- 239000003146 anticoagulant agent Substances 0.000 claims description 115
- 229940127219 anticoagulant drug Drugs 0.000 claims description 115
- 230000018044 dehydration Effects 0.000 claims description 111
- 238000006297 dehydration reaction Methods 0.000 claims description 111
- OYPRJOBELJOOCE-UHFFFAOYSA-N Calcium Chemical compound [Ca] OYPRJOBELJOOCE-UHFFFAOYSA-N 0.000 claims description 109
- 230000017531 blood circulation Effects 0.000 claims description 109
- 229910052791 calcium Inorganic materials 0.000 claims description 109
- DGAQECJNVWCQMB-PUAWFVPOSA-M Ilexoside XXIX Chemical compound C[C@@H]1CC[C@@]2(CC[C@@]3(C(=CC[C@H]4[C@]3(CC[C@@H]5[C@@]4(CC[C@@H](C5(C)C)OS(=O)(=O)[O-])C)C)[C@@H]2[C@]1(C)O)C)C(=O)O[C@H]6[C@@H]([C@H]([C@@H]([C@H](O6)CO)O)O)O.[Na+] DGAQECJNVWCQMB-PUAWFVPOSA-M 0.000 claims description 107
- 229910052708 sodium Inorganic materials 0.000 claims description 107
- ZLMJMSJWJFRBEC-UHFFFAOYSA-N Potassium Chemical compound [K] ZLMJMSJWJFRBEC-UHFFFAOYSA-N 0.000 claims description 101
- 239000011591 potassium Substances 0.000 claims description 101
- 229910052700 potassium Inorganic materials 0.000 claims description 101
- 230000006870 function Effects 0.000 claims description 95
- 238000007637 random forest analysis Methods 0.000 claims description 79
- 230000009467 reduction Effects 0.000 claims description 65
- 230000003442 weekly effect Effects 0.000 claims description 57
- 238000013528 artificial neural network Methods 0.000 claims description 54
- 210000002569 neuron Anatomy 0.000 claims description 50
- 238000005192 partition Methods 0.000 claims description 49
- 238000003066 decision tree Methods 0.000 claims description 40
- 238000007635 classification algorithm Methods 0.000 claims description 32
- 210000002966 serum Anatomy 0.000 claims description 30
- 230000002159 abnormal effect Effects 0.000 claims description 27
- 238000007477 logistic regression Methods 0.000 claims description 22
- 238000012417 linear regression Methods 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 20
- 102000001554 Hemoglobins Human genes 0.000 claims description 18
- 108010054147 Hemoglobins Proteins 0.000 claims description 18
- 230000037396 body weight Effects 0.000 claims description 18
- 210000004369 blood Anatomy 0.000 claims description 17
- 239000008280 blood Substances 0.000 claims description 17
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 claims description 12
- XSQUKJJJFZCRTK-UHFFFAOYSA-N Urea Chemical compound NC(N)=O XSQUKJJJFZCRTK-UHFFFAOYSA-N 0.000 claims description 12
- DDRJAANPRJIHGJ-UHFFFAOYSA-N creatinine Chemical compound CN1CC(=O)NC1=N DDRJAANPRJIHGJ-UHFFFAOYSA-N 0.000 claims description 12
- 210000003743 erythrocyte Anatomy 0.000 claims description 12
- 102000004169 proteins and genes Human genes 0.000 claims description 12
- 108090000623 proteins and genes Proteins 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 10
- 230000036772 blood pressure Effects 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 10
- 230000009466 transformation Effects 0.000 claims description 10
- 238000006243 chemical reaction Methods 0.000 claims description 8
- WQZGKKKJIJFFOK-GASJEMHNSA-N Glucose Natural products OC[C@H]1OC(O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-GASJEMHNSA-N 0.000 claims description 7
- 239000008103 glucose Substances 0.000 claims description 7
- 108010088751 Albumins Proteins 0.000 claims description 6
- 102000009027 Albumins Human genes 0.000 claims description 6
- BVKZGUZCCUSVTD-UHFFFAOYSA-M Bicarbonate Chemical compound OC([O-])=O BVKZGUZCCUSVTD-UHFFFAOYSA-M 0.000 claims description 6
- 102000008857 Ferritin Human genes 0.000 claims description 6
- 108050000784 Ferritin Proteins 0.000 claims description 6
- 238000008416 Ferritin Methods 0.000 claims description 6
- 108010049003 Fibrinogen Proteins 0.000 claims description 6
- 102000008946 Fibrinogen Human genes 0.000 claims description 6
- 102000008133 Iron-Binding Proteins Human genes 0.000 claims description 6
- 108010035210 Iron-Binding Proteins Proteins 0.000 claims description 6
- 102000036675 Myoglobin Human genes 0.000 claims description 6
- 108010062374 Myoglobin Proteins 0.000 claims description 6
- 102000003982 Parathyroid hormone Human genes 0.000 claims description 6
- 108090000445 Parathyroid hormone Proteins 0.000 claims description 6
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 claims description 6
- 108010071690 Prealbumin Proteins 0.000 claims description 6
- 102000007584 Prealbumin Human genes 0.000 claims description 6
- 102000013394 Troponin I Human genes 0.000 claims description 6
- 108010065729 Troponin I Proteins 0.000 claims description 6
- 102000004987 Troponin T Human genes 0.000 claims description 6
- 108090001108 Troponin T Proteins 0.000 claims description 6
- LEHOTFFKMJEONL-UHFFFAOYSA-N Uric Acid Chemical compound N1C(=O)NC(=O)C2=C1NC(=O)N2 LEHOTFFKMJEONL-UHFFFAOYSA-N 0.000 claims description 6
- TVWHNULVHGKJHS-UHFFFAOYSA-N Uric acid Natural products N1C(=O)NC(=O)C2NC(=O)NC21 TVWHNULVHGKJHS-UHFFFAOYSA-N 0.000 claims description 6
- 229940050528 albumin Drugs 0.000 claims description 6
- WQZGKKKJIJFFOK-VFUOTHLCSA-N beta-D-glucose Chemical compound OC[C@H]1O[C@@H](O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-VFUOTHLCSA-N 0.000 claims description 6
- 238000004820 blood count Methods 0.000 claims description 6
- 230000036760 body temperature Effects 0.000 claims description 6
- 239000004202 carbamide Substances 0.000 claims description 6
- 229940109239 creatinine Drugs 0.000 claims description 6
- 229940012952 fibrinogen Drugs 0.000 claims description 6
- 208000015181 infectious disease Diseases 0.000 claims description 6
- 238000007689 inspection Methods 0.000 claims description 6
- 229910052742 iron Inorganic materials 0.000 claims description 6
- 210000000265 leukocyte Anatomy 0.000 claims description 6
- 239000000199 parathyroid hormone Substances 0.000 claims description 6
- 229960001319 parathyroid hormone Drugs 0.000 claims description 6
- 239000011574 phosphorus Substances 0.000 claims description 6
- 229910052698 phosphorus Inorganic materials 0.000 claims description 6
- 229940045136 urea Drugs 0.000 claims description 6
- 229940116269 uric acid Drugs 0.000 claims description 6
- 210000002700 urine Anatomy 0.000 claims description 6
- 238000012423 maintenance Methods 0.000 claims description 5
- VEXZGXHMUGYJMC-UHFFFAOYSA-M Chloride anion Chemical compound [Cl-] VEXZGXHMUGYJMC-UHFFFAOYSA-M 0.000 claims description 4
- 238000004140 cleaning Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 108010058861 Fibrin Fibrinogen Degradation Products Proteins 0.000 claims 2
- 239000000208 fibrin degradation product Substances 0.000 claims 2
- 206010067484 Adverse reaction Diseases 0.000 abstract description 7
- 230000006838 adverse reaction Effects 0.000 abstract description 7
- 102000003951 Erythropoietin Human genes 0.000 description 83
- 108090000394 Erythropoietin Proteins 0.000 description 83
- 229940105423 erythropoietin Drugs 0.000 description 83
- OXCMYAYHXIHQOA-UHFFFAOYSA-N potassium;[2-butyl-5-chloro-3-[[4-[2-(1,2,4-triaza-3-azanidacyclopenta-1,4-dien-5-yl)phenyl]phenyl]methyl]imidazol-4-yl]methanol Chemical compound [K+].CCCCC1=NC(Cl)=C(CO)N1CC1=CC=C(C=2C(=CC=CC=2)C2=N[N-]N=N2)C=C1 OXCMYAYHXIHQOA-UHFFFAOYSA-N 0.000 description 83
- 230000000875 corresponding effect Effects 0.000 description 19
- 229940079593 drug Drugs 0.000 description 5
- 102000009123 Fibrin Human genes 0.000 description 4
- 108010073385 Fibrin Proteins 0.000 description 4
- BWGVNKXGVNDBDI-UHFFFAOYSA-N Fibrin monomer Chemical compound CNC(=O)CNC(=O)CN BWGVNKXGVNDBDI-UHFFFAOYSA-N 0.000 description 4
- 239000007857 degradation product Substances 0.000 description 4
- 229950003499 fibrin Drugs 0.000 description 4
- 239000012530 fluid Substances 0.000 description 4
- 229940099990 ogen Drugs 0.000 description 4
- 208000013016 Hypoglycemia Diseases 0.000 description 3
- 230000002596 correlated effect Effects 0.000 description 3
- 239000000385 dialysis solution Substances 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000010348 incorporation Methods 0.000 description 3
- 230000002093 peripheral effect Effects 0.000 description 3
- 206010005746 Blood pressure fluctuation Diseases 0.000 description 2
- ZAMOUSCENKQFHK-UHFFFAOYSA-N Chlorine atom Chemical compound [Cl] ZAMOUSCENKQFHK-UHFFFAOYSA-N 0.000 description 2
- 206010020772 Hypertension Diseases 0.000 description 2
- 208000001953 Hypotension Diseases 0.000 description 2
- 239000000460 chlorine Substances 0.000 description 2
- 229910052801 chlorine Inorganic materials 0.000 description 2
- 230000001112 coagulating effect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 239000000004 hemodialysis solution Substances 0.000 description 2
- 230000002218 hypoglycaemic effect Effects 0.000 description 2
- 230000036543 hypotension Effects 0.000 description 2
- 230000008676 import Effects 0.000 description 2
- 239000012528 membrane Substances 0.000 description 2
- 239000002243 precursor Substances 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000000108 ultra-filtration Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 208000012639 Balance disease Diseases 0.000 description 1
- 208000024172 Cardiovascular disease Diseases 0.000 description 1
- 206010013954 Dysphoria Diseases 0.000 description 1
- 206010062713 Haemorrhagic diathesis Diseases 0.000 description 1
- SNIOPGDIGTZGOP-UHFFFAOYSA-N Nitroglycerin Chemical compound [O-][N+](=O)OCC(O[N+]([O-])=O)CO[N+]([O-])=O SNIOPGDIGTZGOP-UHFFFAOYSA-N 0.000 description 1
- 239000000006 Nitroglycerin Substances 0.000 description 1
- 206010033557 Palpitations Diseases 0.000 description 1
- FAPWRFPIFSIZLT-UHFFFAOYSA-M Sodium chloride Chemical compound [Na+].[Cl-] FAPWRFPIFSIZLT-UHFFFAOYSA-M 0.000 description 1
- 208000005392 Spasm Diseases 0.000 description 1
- 206010047700 Vomiting Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 208000020832 chronic kidney disease Diseases 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013501 data transformation Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 208000002173 dizziness Diseases 0.000 description 1
- 230000002526 effect on cardiovascular system Effects 0.000 description 1
- 239000003792 electrolyte Substances 0.000 description 1
- 208000028208 end stage renal disease Diseases 0.000 description 1
- 201000000523 end stage renal failure Diseases 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 229960003711 glyceryl trinitrate Drugs 0.000 description 1
- 208000031169 hemorrhagic disease Diseases 0.000 description 1
- 238000001727 in vivo Methods 0.000 description 1
- 238000001802 infusion Methods 0.000 description 1
- 238000011221 initial treatment Methods 0.000 description 1
- 210000003734 kidney Anatomy 0.000 description 1
- 208000017169 kidney disease Diseases 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003204 osmotic effect Effects 0.000 description 1
- 208000037920 primary disease Diseases 0.000 description 1
- 238000000746 purification Methods 0.000 description 1
- 238000009256 replacement therapy Methods 0.000 description 1
- 239000011780 sodium chloride Substances 0.000 description 1
- 238000001179 sorption measurement Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000035900 sweating Effects 0.000 description 1
- 239000003053 toxin Substances 0.000 description 1
- 231100000765 toxin Toxicity 0.000 description 1
- 108700012359 toxins Proteins 0.000 description 1
- 238000011282 treatment Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/40—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to mechanical, radiation or invasive therapies, e.g. surgery, laser therapy, dialysis or acupuncture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H70/00—ICT specially adapted for the handling or processing of medical references
- G16H70/40—ICT specially adapted for the handling or processing of medical references relating to drugs, e.g. their side effects or intended usage
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Epidemiology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Surgery (AREA)
- Bioinformatics & Computational Biology (AREA)
- Toxicology (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Pharmacology & Pharmacy (AREA)
- Algebra (AREA)
- Medicinal Chemistry (AREA)
- Databases & Information Systems (AREA)
- Urology & Nephrology (AREA)
- Evolutionary Biology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
Abstract
The invention provides an AI hemodialysis scheme model modeling method and a method for generating a hemodialysis scheme, which comprise a dialyzer submodel, a dialysis setting submodel, a dialysis medicament submodel and a dialysate submodel; using the conclusion of the dialyzer submodel as an input variable to be applied to the dialysis setting submodel; using the conclusion of the dialysis setting submodel as an input variable to be applied to the dialysis medicament submodel; applying the conclusion of the dialysis agent submodel as an input variable to the dialysate submodel to finally provide a personalized hemodialysis scheme; not only can assist medical personnel to formulate a personalized dialysis scheme, but also can reduce the labor intensity of the medical personnel and the professional requirement; the risk of adverse reaction in dialysis can be reduced or avoided, the safety of patients is ensured, and the medical care quality is improved; and moreover, the prediction error can be reduced, and the accuracy of the prediction result can be improved.
Description
Technical Field
The invention relates to the technical field of blood purification, in particular to an AI hemodialysis scheme model modeling method, and a method, a system and an application for generating a hemodialysis scheme by an AI hemodialysis scheme model created based on the AI hemodialysis scheme model modeling method.
Background
Maintenance hemodialysis, a kidney replacement therapy, is the primary treatment for end-stage renal disease, and about 90% of dialysis patients require maintenance dialysis treatment. Hemodialysis is to introduce the blood of a patient and dialysate into a dialyzer at the same time, flow in opposite directions on two sides of a dialysis membrane, and remove toxins through dispersion, convection and adsorption by virtue of solubility gradient, osmotic gradient and water pressure gradient on two sides of a semipermeable membrane; removing excessive water in vivo by ultrafiltration and osmosis; meanwhile, the required substances are supplemented, and electrolyte and acid-base balance disorder is corrected. Hemodialysis is a medical care action with strong specialty and great risk. The patient has dialysis risk in the dialysis period, adverse reactions (adjustment of anticoagulant, blood pressure fluctuation, hypotension, hypertension, hypoglycemia, cardiovascular disease and other abnormal conditions) often occur, a perfect personalized dialysis scheme is worked out, and the dialysis risk can be effectively solved to the minimum, so that the safety of the patient is ensured, and the medical care quality is improved.
At present, in a dialysis period, most of the dialysis solutions are prepared by the experience of medical staff, but the individual conditions of different patients are complex, and the experience and energy of the medical staff are limited. In addition, during dialysis, the detection of vital signs is performed by the detection of specialized instruments. The vital signs are monitored once every certain time (such as half an hour or so) by using the instrument, the workload is increased due to frequent operation, and the possibility of emergency situations is increased due to non-real-time monitoring. There is a need to improve the prior art to provide a more sophisticated personalized dialysis solution that relieves the pressure on the operation of medical personnel and specialized equipment.
Disclosure of Invention
In order to solve the technical problem, the invention provides a method for constructing an AI hemodialysis scheme generation model, wherein the AI hemodialysis scheme generation model comprises a dialyzer submodel, a dialysis setting submodel, a dialysis medicament submodel and a dialysate submodel; using the conclusion of the dialyzer submodel as an input variable for a dialysis setting submodel, using the conclusion of the dialysis setting submodel as an input variable for a dialysis agent submodel, and using the conclusion of the dialysis agent submodel as an input variable for a dialysate submodel; the dialyzer model gives dialyzer model parameter values; a dialysis setting submodel gives a preset dehydration value parameter; the dialysis agent sub-model gives anticoagulant dosage parameters and EPO week dosage parameters; the dialysate submodel gives parameters of dialysate flow, dialysate temperature, conductivity, potassium concentration, sodium concentration and calcium concentration. In a further aspect of the present invention,
a dialyzer submodel; the dialyzer submodel comprises a dialyzer model component and a dialysis mode component;
the dialyzer model component comprises a dialyzer model algorithm component and a dialyzer model feature importance analysis component; the dialyzer model component gives the dialyzer model component input characteristics through the dialyzer model characteristic importance degree analysis component, inputs the dialyzer model component input characteristics into the dialyzer model algorithm component, and finally gives dialyzer model parameter values through the dialyzer model algorithm component;
by the input characteristics given by the dialyzer model characteristic importance degree analysis component, on one hand, the number of the characteristics can be effectively reduced, and thus the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and the difficulty of a learning task of a dialyzer model algorithm component is reduced.
Furthermore, the dialyzer model feature importance analysis component analyzes the feature importance of the input features by using an xgboost algorithm, fits all the input features and the actual dialyzer model by using xgboost, sorts the average reduction of the loss function when each input feature is used as a partition attribute from large to small to obtain a feature importance ranking, and uses the input feature with the top ranking as the dialyzer model algorithm component input feature.
The xgboost algorithm, through the gradient boosting method, can obtain the importance of each feature according to the tree after boosting. Therefore, the characteristic of low importance is removed, and the robustness of the dialyzer model algorithm component is enhanced.
The dialyzer model algorithm component is composed of a decision classification tree, the decision classification tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold of the input feature, the input features are classified layer by layer according to the thresholds, and finally the input features are classified into a certain dialyzer category;
and aiming at the dialyzer model algorithm component, the dialyzer category is predicted by using the decision classification tree, so that the prediction accuracy can be improved.
The dialysis mode component comprises a dialysis mode algorithm component and a dialysis mode feature importance analysis component; the dialysis mode component firstly gives the input characteristics of the dialysis mode component through the dialysis mode characteristic importance degree analysis component, and then inputs the input characteristics of the dialysis mode component into the dialysis mode algorithm component; finally, the dialysis mode algorithm component gives the dialysis mode parameter values.
By the input characteristics given by the dialysis mode characteristic importance degree analysis component, on one hand, the number of the characteristics can be effectively reduced, and the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and the difficulty of a dialysis mode algorithm component learning task is reduced.
Furthermore, the dialysis mode feature importance degree analysis module performs feature importance degree analysis on the input features by using an xgboost algorithm, fits all the input features and the dialysis modes by using an xgboost, ranks the average reduction amount of the loss function when each input feature is used as a partition attribute from large to small to obtain a feature importance degree rank, and uses the input feature ranked at the top as the input feature of the dialysis mode algorithm module.
The xgboost algorithm, by the gradient boosting method, can obtain the importance of each feature according to the tree after boosting. Therefore, the characteristic with low importance is eliminated, and the robustness of the dialysis mode algorithm component is enhanced.
The dialysis mode algorithm component is composed of a logistic regression algorithm, and firstly, input features are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysis mode; the dialysis mode with the maximum probability is the output result of the dialysis mode algorithm component;
and aiming at the dialysis mode algorithm component, the classification of the dialysis mode is predicted by using a logistic regression algorithm, so that the prediction accuracy is improved.
The dialyzer model parameter values and dialysis mode parameter values given by the dialyzer submodel constitute the most basic content in the fluid dialysis protocol.
A dialysis setting sub-model; the dialysis setting submodel comprises a preset dehydration component, a dialysis time component and a dialysis blood flow component;
the preset dehydration component comprises a preset dehydration algorithm component and a preset dehydration characteristic importance degree analysis component; the preset dehydration component firstly gives the input characteristics of the preset dehydration component through the preset dehydration characteristic importance analysis component; then, inputting the input characteristics of the preset dehydration component into a preset dehydration algorithm component; finally, a preset dehydration algorithm component gives a preset dehydration value parameter;
by presetting the input characteristics given by the dehydration characteristic importance degree analysis component, on one hand, the number of the characteristics can be effectively reduced, and the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and the difficulty of the preset dehydration algorithm component learning task is reduced.
Further, the preset dehydration characteristic importance degree analysis component analyzes the input characteristics by using an xgboost algorithm, fits all the input characteristics and the actual preset dehydration value by using the xgboost, sorts the average reduction quantity of the loss function when each characteristic is used as a partition attribute from large to small to obtain characteristic importance degree sorting, and takes the input characteristic with the top ranking as the input characteristic of the preset dehydration algorithm component;
the xgboost algorithm can acquire the importance of each feature according to the promoted tree by a gradient promotion method, so that the feature with low importance is removed, and the robustness of the preset dehydration algorithm component is enhanced.
The preset dehydration algorithm component is composed of a BP neural network regression algorithm, the BP neural network regression algorithm is composed of a plurality of layers of hidden layers and an output layer, each layer of hidden layers and the output layer are composed of a plurality of neurons, each neuron in the first hidden layer combines input features in a multiple regression equation mode, an equation result is subjected to nonlinear transformation through an activation function, then the neuron of each hidden layer performs multiple regression on data obtained by the previous layer, the equation result is subjected to nonlinear transformation through the activation function, the number of the neurons of the output layer is 1, only the previous numerical value is subjected to multiple linear regression, and the final value output by the multiple linear regression is the preset dehydration value predicted by the model.
The applicant has tried various methods for presetting the dehydration algorithm components. The preset dehydration value is predicted by using the BP neural network regression algorithm, the accuracy rate is highest, and therefore the BP neural network regression algorithm is selected to form a preset dehydration algorithm component.
The dialysis time component comprises a dialysis time algorithm component and a dialysis time characteristic importance degree analysis component; the dialysis time component firstly gives the input characteristics of the dialysis time component through the dialysis time characteristic importance degree analysis component; then, inputting the input characteristics of the dialysis time component into a dialysis time algorithm component; finally, a dialysis time algorithm component gives the dialysis time parameter.
On one hand, the number of the features can be effectively reduced through the input features given by the dialysis time feature importance analysis component, so that the problem of excessive features is solved; on the other hand, irrelevant features are removed, only key features are left, and difficulty of a dialysis time algorithm component learning task is reduced.
Further, the dialysis time feature importance degree analysis module performs feature importance degree analysis on the input features by using an xgboost algorithm, fits all the input features and the dialysis time values by using the xgboost, sorts the average reduction amount of the loss function of each feature when the feature is used as a partition attribute from large to small to obtain a feature importance degree sort, and uses the input feature with the top rank as the input feature of the dialysis time algorithm module;
the method uses the xgboost algorithm, and can acquire the importance of each feature according to the boosted tree through a gradient boosting method. Therefore, the characteristic with low importance is removed, and the robustness of the dialysis time algorithm component is enhanced.
The dialysis time algorithm component is composed of classified random forests, each random forest is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the random forest is the dialysis time parameter.
Aiming at the dialysis time algorithm component, the invention predicts the type of the dialysis time by using the random forest, thereby improving the accuracy of prediction.
The dialysis blood flow component comprises a dialysis blood flow algorithm component and a dialysis blood flow characteristic importance analysis component; firstly, the dialysis blood flow algorithm component gives the input characteristics of the dialysis blood flow component through a dialysis blood flow characteristic importance degree analysis component; then, inputting the input characteristics of the dialysis blood flow component into a dialysis blood flow algorithm component; finally, a dialysis blood flow algorithm component gives a dialysis blood flow parameter.
According to the invention, the input characteristics given by the dialysis blood flow characteristic importance degree analysis component can effectively reduce the number of characteristics on one hand, so that the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and the difficulty of the learning task of the dialysis blood flow algorithm component is reduced.
Furthermore, the dialysis blood flow characteristic importance degree analysis module analyzes the characteristic importance degree of the input characteristics by an xgboost algorithm, fits all the input characteristics and the dialysis blood flow value by the xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristic is used as a division attribute from large to small to obtain characteristic importance degree sorting, and uses the input characteristic with the top ranking as the input characteristic of the dialysis blood flow algorithm module.
According to the method, the xgboost algorithm is adopted, and the importance of each feature can be obtained according to the promoted tree through a gradient promotion method, so that the features with low importance are eliminated, and the robustness of a blood flow algorithm component is enhanced.
The blood flow algorithm component is composed of a multiple linear regression equation, and the input characteristics are substituted into the multiple linear regression equation to obtain a final numerical value, namely the value of the blood flow parameter.
Aiming at the blood flow algorithm component, the blood flow value is predicted by using the multiple linear regression equation, and the prediction accuracy is high.
A dialysis agent sub-model; the dialysis agent sub-model comprises an anticoagulant component and an EPO weekly dose component;
the anticoagulant component comprises an anticoagulant algorithm component and an anticoagulant characteristic importance analysis component;
the anticoagulant component firstly gives the input characteristics of the anticoagulant component through the anticoagulant characteristic importance analysis component; then inputting the input characteristics of the anticoagulant component into the anticoagulant algorithm component; finally, the anticoagulant algorithm component gives anticoagulant dosage parameters;
according to the input characteristics given by the anticoagulant characteristic importance analysis component, on one hand, the number of the characteristics can be effectively reduced, and therefore the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and difficulty of a learning task of the anticoagulant algorithm component is reduced.
Further, the anticoagulant feature importance degree analysis module performs feature importance degree analysis on the input features by using an xgboost algorithm, all the input features and the anticoagulant dose are fitted by using the xgboost, the average reduction of the loss function of each feature when the feature is used as a partition attribute is ranked from large to small to obtain a feature importance degree ranking, and the input feature ranked at the top is used as the input feature of the anticoagulant algorithm module.
The method adopts an xgboost algorithm, and can acquire the importance of each feature according to the boosted tree by a gradient boosting method. Therefore, the characteristic of low importance is eliminated, and the robustness of the anticoagulant dose algorithm component is enhanced.
The anticoagulant dose algorithm component is composed of a decision regression tree algorithm, the decision regression tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold value of the input feature, the nodes are divided layer by layer according to the threshold values, the final output result is a mean value of numerical values in the node, and the mean value of the numerical values in the node is the anticoagulant dose.
Aiming at the anticoagulant dose algorithm component, the decision regression tree is utilized to predict the anticoagulant dose, so that the accuracy of prediction is improved, and the decision regression tree is selected to form the anticoagulant dose algorithm component.
The EPO week dosage component comprises an EPO week dosage algorithm component and an EPO week dosage characteristic importance analysis component; the EPO week dosage component firstly gives input characteristics of the EPO week dosage component through the EPO week dosage characteristic importance analysis component; then, the EPO weekly dose component input features are input into the EPO weekly dose algorithm component; finally, the EPO weekly dose algorithm component gives the EPO weekly dose parameters.
According to the input characteristics given by the EPO week dose characteristic importance analysis component, on one hand, the number of the characteristics can be effectively reduced, so that the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and the difficulty of the learning task of the EPO weekly dose algorithm component is reduced.
Further, the characteristic importance degree analysis module of the peripheral EPO dose analyzes the characteristic importance degree of the input characteristics by an xgboost algorithm, fits all the input characteristics and the peripheral EPO dose by the xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristic is used as a division attribute from large to small to obtain a characteristic importance degree sequence, and uses the input characteristic with the top rank as the input characteristic of the peripheral EPO dose algorithm module.
The importance of each feature can be obtained according to the tree after lifting by adopting an xgboost algorithm and a gradient lifting method. Therefore, the characteristic of low importance is eliminated, and the robustness of the EPO week dosage algorithm component is enhanced.
Further, the EPO refers to a drug intended for use in a dialysis regimen; the EPO weekly dose refers to the dose of a planned week of administration in a dialysis regimen.
The EPO weekly dose algorithm component is composed of a decision regression tree, the decision regression tree is composed of a plurality of nodes, each node is provided with an input characteristic and a threshold of the input characteristic, the nodes are divided layer by layer according to the threshold, the final output result is the mean value of numerical values in the nodes, and the mean value of the numerical values in the nodes is the EPO weekly dose.
Aiming at the EPO weekly dosage algorithm component, the invention predicts the EPO weekly dosage by using a decision regression tree, thereby improving the prediction accuracy.
A dialysate sub-model; the dialysate submodel comprises a dialysate flow component, a dialysate temperature component, a dialysate conductivity component, a dialysate potassium concentration component, a dialysate sodium concentration component and a dialysate calcium concentration component;
the dialysate flow component comprises a dialysate flow algorithm component and a dialysate flow characteristic importance analysis component; firstly, the dialysate flow component gives input characteristics of the dialysate flow component through a dialysate flow characteristic importance analysis component; then, inputting the dialysate flow component input characteristics into a dialysate flow algorithm component; finally, the dialysate flow algorithm component gives dialysate flow parameters;
according to the invention, the input characteristics given by the dialysate flow characteristic importance degree analysis component are utilized, on one hand, the number of characteristics can be effectively reduced, and thus the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and difficulty of a dialysate flow algorithm component learning task is reduced.
Furthermore, the dialysate flow characteristic importance analysis module performs characteristic importance analysis on the input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate flow by using the xgboost, sorts the average reduction of the loss function of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance ranking, and uses the input characteristic ranked at the top as the input characteristic of the dialysate flow algorithm module.
The importance of each feature can be obtained according to the tree after lifting by adopting an xgboost algorithm and a gradient lifting method. Therefore, the characteristic of low importance is eliminated, and the robustness of the dialysate flow algorithm component is enhanced.
The dialysate flow algorithm component is composed of a logistic regression algorithm, and firstly, input characteristics are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysate flow; the dialysate flow with the maximum probability is the output result of the dialysate flow algorithm component;
for the dialysate flow algorithm component, the invention predicts the dialysate flow by using logistic regression, and has higher prediction accuracy.
The dialysate temperature component comprises a dialysate temperature algorithm component and a dialysate temperature characteristic importance analysis component; the temperature component of the dialysate firstly gives the input characteristics of the temperature component of the dialysate through the temperature characteristic importance degree analysis component of the dialysate; then inputting the dialysate temperature component input characteristics into a dialysate temperature algorithm component; finally, the dialysate temperature algorithm component gives a dialysate temperature parameter;
according to the method, the input characteristics given by the dialysate temperature characteristic importance degree analysis component are utilized, on one hand, the number of the characteristics can be effectively reduced, and therefore the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and difficulty of a learning task of the dialysate temperature algorithm component is reduced.
Further, the dialysate temperature feature importance analyzing module analyzes the feature importance of the input features by using an xgboost algorithm, fits all the input features and the dialysate temperature by using an xgboost, sorts the average reduction of the loss function of each feature as a partition attribute from large to small to obtain a feature importance ranking, and takes the input feature ranked at the front as the input feature of the dialysate temperature algorithm module;
the importance of each feature can be obtained according to the tree after lifting by adopting an xgboost algorithm and a gradient lifting method. Therefore, the characteristic of low importance is removed, and the robustness of the dialysate temperature algorithm component is enhanced.
The dialysate temperature algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate temperature parameter.
Aiming at the dialysate temperature algorithm component, the method utilizes the classified random forest to predict the dialysate temperature, and the prediction accuracy is high.
The dialysate conductivity component comprises a dialysate conductivity algorithm component and a dialysate conductivity feature importance analysis component; firstly, the dialysate conductivity component gives input characteristics of the dialysate conductivity component through the dialysate conductivity characteristic importance analysis component; then inputting the input characteristics of the dialysate conductivity component into a dialysate conductivity algorithm component; finally, the dialysate conductivity algorithm component gives parameters of the conductivity of the dialysate;
according to the invention, the input characteristics given by the dialysate conductivity characteristic importance degree analysis component are utilized, on one hand, the number of characteristics can be effectively reduced, and thus the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and difficulty of a dialysis fluid conductivity algorithm component learning task is reduced.
Furthermore, the dialysate conductivity feature importance analysis module analyzes the feature importance of the input features by using an xgboost algorithm, fits all the input features and the dialysate conductivity by using an xgboost, sorts the average reduction of loss functions of each feature as a division attribute from large to small to obtain a feature importance ranking, and takes the input feature ranked at the top as the input feature of the dialysate conductivity algorithm module.
The method adopts an xgboost algorithm, and can acquire the importance of each feature according to the boosted tree by a gradient boosting method. Therefore, the characteristic of low importance is removed, and the robustness of the dialysate conductivity algorithm component is enhanced.
The dialysate conductivity algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely, the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the conductivity parameter of the dialysate.
Aiming at the dialysate conductivity algorithm component, the invention predicts the conductivity of the dialysate by using the classified random forest, and has high accuracy.
The dialysate potassium concentration component comprises a dialysate potassium concentration algorithm component and a dialysate potassium concentration characteristic importance analysis component; firstly, the dialysate potassium concentration component gives input characteristics of the dialysate potassium concentration component through a dialysate potassium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate potassium concentration component into a dialysate potassium concentration algorithm component; finally, a dialysate potassium concentration algorithm component gives a dialysate potassium concentration parameter;
according to the method, the input characteristics given by the dialysate potassium concentration characteristic importance analysis component are used, so that on one hand, the number of the characteristics can be effectively reduced, and the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and difficulty of a learning task of a potassium concentration algorithm component of the dialysate is reduced.
Furthermore, the dialysate potassium concentration characteristic importance degree analysis module performs characteristic importance degree analysis on the input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate potassium concentration by using the xgboost, sorts the average reduction amount of the loss function of each characteristic from large to small when the characteristics are used as the partition attributes to obtain a characteristic importance degree sort, and uses the input characteristic ranked at the top as the input characteristic of the dialysate potassium concentration algorithm module.
The importance of each feature can be obtained according to the tree after lifting by adopting an xgboost algorithm and a gradient lifting method. Therefore, the characteristic of low importance is eliminated, and the robustness of the dialysate potassium concentration algorithm component is enhanced.
The dialysate potassium concentration algorithm component is composed of a classified random forest which is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate potassium concentration parameter.
Aiming at the potassium concentration algorithm component of the dialysate, the invention predicts the potassium concentration of the dialysate by utilizing the classified random forest, and has high accuracy.
The dialysate sodium concentration component comprises a dialysate sodium concentration algorithm component and a dialysate sodium concentration characteristic importance analysis component; firstly, the dialysate sodium concentration component gives input characteristics of the dialysate sodium concentration component through a dialysate sodium concentration characteristic importance analysis component; then inputting the input characteristics of the dialysate sodium concentration component into a dialysate sodium concentration algorithm component; finally, a dialysate sodium concentration algorithm component gives a dialysate sodium concentration parameter;
according to the method, the input characteristics given by the dialysate sodium concentration characteristic importance analysis component are used, so that on one hand, the number of the characteristics can be effectively reduced, and the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and difficulty of a learning task of a dialysate sodium concentration algorithm component is reduced.
Furthermore, the dialysate sodium concentration characteristic importance analysis module performs characteristic importance analysis on the input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate sodium concentration by using an xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristic is used as a division attribute from large to small to obtain characteristic importance sorting, and uses the input characteristic with the top ranking as the input characteristic of the dialysate sodium concentration algorithm module.
The method adopts an xgboost algorithm, and can acquire the importance of each feature according to the boosted tree by a gradient boosting method. Therefore, the characteristic of low importance is eliminated, and the robustness of the dialysate sodium concentration algorithm component is enhanced.
The dialysate sodium concentration algorithm component is composed of a BP neural network classification algorithm; the other structures of the BP neural network classification algorithm except an output layer are the same as the BP neural network regression algorithm, the output layer of the BP neural network classification algorithm is the classification category number of dialysate sodium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of the hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to the range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate sodium concentration category is obtained.
Aiming at the dialysate sodium concentration algorithm component, the invention predicts the dialysate sodium concentration by using a BP neural network classification algorithm, and has high accuracy.
The dialysate calcium concentration component comprises a dialysate calcium concentration algorithm component and a dialysate calcium concentration characteristic importance analysis component; firstly, the dialysate calcium concentration component gives the input characteristics of the dialysate calcium concentration component through a dialysate calcium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate calcium concentration component into a dialysate calcium concentration algorithm component; finally, the dialysate calcium concentration algorithm component provides a dialysate calcium concentration parameter.
According to the method, the input characteristics given by the dialysate calcium concentration characteristic importance analysis component are utilized, so that on one hand, the number of characteristics can be effectively reduced, and the problem of excessive characteristics is solved; on the other hand, irrelevant features are removed, only key features are left, and difficulty of a learning task of a dialysate calcium concentration algorithm component is reduced.
Furthermore, the dialysate calcium concentration feature importance analysis module performs feature importance analysis on the input features by using an xgboost algorithm, fits all the input features and the dialysate calcium concentration by using an xgboost, sorts the average reduction amount of the loss function of each feature as a division attribute from large to small to obtain a feature importance ranking, and uses the input feature ranked at the top as the input feature of the dialysate calcium concentration algorithm module.
The method adopts an xgboost algorithm, and can acquire the importance of each feature according to the boosted tree by a gradient boosting method. Therefore, the characteristic of low importance is removed, and the robustness of the dialysate calcium concentration algorithm component is enhanced.
The dialysate calcium concentration algorithm component is composed of a BP neural network classification algorithm; the other structures of the BP neural network classification algorithm except an output layer are the same as the BP neural network regression algorithm, the output layer of the BP neural network classification algorithm is the number of classification categories of dialysate calcium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of the hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to the range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate calcium concentration category is obtained.
Aiming at the dialysate calcium concentration algorithm component, the invention adopts BP neural network classification algorithm to predict the dialysate calcium concentration, and the prediction accuracy is high.
Finally, the AI hemodialysis plan generating model combines the dialysis mode parameter value, the dialyzer model parameter value, the pre-dehydration value parameter, the dialysis time parameter, the dialysis blood flow parameter, the anticoagulant dosage parameter, the EPO weekly dosage parameter, the dialysate flow parameter, the dialysate temperature parameter, the dialysate conductivity parameter, the dialysate potassium concentration parameter, the dialysate sodium concentration parameter and the dialysate calcium concentration parameter to generate the hemodialysis plan.
The invention provides a method for generating a hemodialysis solution based on a model created by the AI hemodialysis solution generation model modeling method, which comprises the following steps of:
s1, inputting the primary input characteristics into a dialyzer submodel, wherein the dialyzer submodel gives dialysis mode parameter values and dialyzer model parameter values;
the primary input features are derived from a hemodialysis database of a medical institution, and comprise basic information, pre-dialysis grid examination, historical dialysis schemes and historical examination and examination data;
s2, combining the input characteristics with the dialysis mode parameter values and the dialyzer model parameter values to generate secondary input characteristics;
s3, inputting the secondary input characteristics into a dialysis setting submodel, and outputting a pre-dehydration value parameter, a dialysis time parameter and a dialysis blood flow parameter by the dialysis setting submodel;
s4, combining the primary input features with the pre-dehydration value parameters, the dialysis time parameters and the dialysis blood flow volume parameters to generate tertiary input features;
s5, inputting the three-time input characteristics into a dialysis agent sub-model, and giving anticoagulant dose parameters and EPO (erythropoietin) weekly dose parameters by the dialysis agent sub-model;
s6, combining the primary input characteristics with anticoagulant dose parameters and EPO week dose parameters to generate four-time input characteristics;
s7, inputting the four input characteristics into a dialysate submodel, wherein the dialysate submodel gives a dialysate flow parameter, a dialysate temperature parameter, a dialysate conductivity parameter, a dialysate potassium concentration parameter, a dialysate sodium concentration parameter and a dialysate calcium concentration parameter;
and S8, combining the dialysis mode parameter value, the dialyzer model parameter value, the pre-dehydration value parameter, the dialysis time parameter, the dialysis blood flow parameter, the anticoagulant dosage parameter, the EPO week dosage parameter, the dialysate flow parameter, the dialysate temperature parameter, the dialysate conductivity parameter, the dialysate potassium concentration parameter, the dialysate sodium concentration parameter and the dialysate calcium concentration parameter to generate a hemodialysis scheme.
Further, the basic information data comprises age, sex, height, weight, blood type, nationality, dialysis age, primary disease, infection history and important medical history;
the pre-dialysis lattice examination comprises dry body weight, last off-machine body weight, pre-dialysis body weight, blood pressure (low pressure), blood pressure (high pressure), heart rate and body temperature;
the historical dialysis scheme comprises 13 indexes including a dialysis mode, a dialyzer, preset dehydration, dialysis time, anticoagulant dose, blood flow, EPO weekly dose, dialysate K concentration, dialysate Ca concentration, dialysate Na concentration, dialysate flow, conductivity and dialysate temperature;
the historical test examination data includes hemoglobin, mean hemoglobin concentration, mean hemoglobin content, platelet count, white blood cell count, red blood cell specific volume, mean red blood cell volume, creatinine, urea, uric acid, albumin, total protein, prealbumin, glucose, sodium, potassium, calcium, phosphorus, chlorine, bicarbonate, plasma fibrinogen, fibrin (ogen) degradation products, plasma D-D dimer, serum troponin I, serum troponin T, myoglobin, serum iron, serum total iron binding capacity, serum ferritin, parathyroid hormone, occult blood, urine protein.
Further, step S1 further comprises a sub-step of giving an input feature, comprising:
s11, collecting characteristic data;
acquiring existing hemodialysis overall process data in a hemodialysis database of a medical institution, and selecting dialysis data of a maintenance dialysis patient, wherein the dialysis data comprises basic information data, dialysis precursor grid inspection data, historical dialysis scheme data and historical inspection data;
the basic information data comprises age, sex, height, weight, blood type, nationality, dialysis age, primary morbidity, infection history and important medical history;
the pre-dialysis check data comprises dry body weight, last off-line body weight, pre-dialysis body weight, blood pressure (low pressure), blood pressure (high pressure), heart rate, body temperature;
the historical dialysis scheme data comprises 13 indexes including a dialysis mode, a dialyzer, preset dehydration, dialysis time, anticoagulant dose, blood flow, EPO weekly dose, dialysate K concentration, dialysate Ca concentration, dialysate Na concentration, dialysate flow, conductivity and dialysate temperature;
the historical test examination data includes hemoglobin, mean hemoglobin concentration, mean hemoglobin content, platelet count, white blood cell count, red blood cell specific volume, mean red blood cell volume, creatinine, urea, uric acid, albumin, total protein, prealbumin, glucose, sodium, potassium, calcium, phosphorus, chloride, bicarbonate, plasma fibrinogen, fibrin (ogen) degradation products, plasma D-D dimer, serum troponin I, serum troponin T, myoglobin, serum iron, serum total iron binding capacity, serum ferritin, parathyroid hormone, occult blood, urine protein.
S12, cleaning the characteristic data to obtain standardized characteristic data; the characteristic data comprises numerical characteristic data and non-numerical characteristic data;
s121, performing unit conversion on each numerical characteristic data of each type, unifying units of each numerical characteristic data of each type, and converting the unified units of each numerical characteristic data into numerical characteristic data of unified units;
s122, processing abnormal values in numerical characteristic data of the unified unit;
the abnormal value processing method comprises the following steps:
s1221, if the numerical data in the numerical characteristic data of the unified unit obeys normal distribution, the data in the numerical characteristic data and the data which is more than or less than 3 times of the standard deviation from the average value of the data in the numerical characteristic data are abnormal values; then, eliminating the data of the row where the abnormal value is located; the average value is the average value of the columns where the numerical characteristic data are located;
s1222, if the numerical characteristics in the numerical characteristic data of the unified unit do not obey normal distribution, adopting a box-line graph method; the numerical characteristic data of the unified unit is smaller than Q1-1.5QR, or the numerical characteristic data of the unified unit is larger than Q3+1.5QR to be an abnormal value. Then the data of the row where the abnormal value is located is culled.
Further, Q1 is the lower quartile, Q3 is the upper quartile, and QR is the subtraction of the lower quartile from the upper quartile.
Obtaining numerical characteristic data of a unified unit with abnormal values removed;
s123, synchronization, namely assigning the non-numerical characteristic data according to a non-numerical data conversion rule to obtain the assigned non-numerical characteristic data;
example (c): the non-numerical data conversion rule is shown in FIG. 12
And S124, finally, forming standardized feature data by the numerical feature data of the unified unit with the abnormal values removed and the assigned non-numerical feature data.
The normalized feature data is a primary input feature.
Has the beneficial effects that: the AI hemodialysis scheme generation model and the method for generating the hemodialysis scheme by the AI hemodialysis scheme generation model can assist medical workers to formulate personalized dialysis schemes, and reduce the labor intensity and professional requirements of the medical workers. Secondly, the risk of dialysis can be reduced through prediction and evaluation, the risk of adverse reaction in dialysis is reduced or avoided, the safety of patients is ensured, and the quality of medical care is improved. Thirdly, the AI hemodialysis scheme generation model is used for generating the hemodialysis scheme, and the conclusion of the dialyzer submodel is used as an input variable for the dialysis setting submodel; using the conclusion of the dialysis setting submodel as an input variable to be applied to the dialysis medicament submodel; using the conclusion of the dialysis agent submodel as an input variable to be applied to the dialysate submodel; the method can reduce the prediction errors of the pre-dehydration value parameter, the dialysis time parameter, the dialysis blood flow parameter, the anticoagulant dosage parameter, the EPO week dosage parameter, the dialysate flow parameter, the dialysate temperature parameter, the dialysate conductivity parameter, the dialysate potassium concentration parameter, the dialysate sodium concentration parameter and the dialysate calcium concentration parameter, and improve the accuracy of the prediction result. Fourthly, the input features which are ranked at the top are used as the input features of each component, particularly the input features of the first fifteen are used as the input features of each component, so that on one hand, inaccuracy of a prediction result caused by incomplete medical feature data is avoided, on the other hand, important medical features can be highlighted, and the accuracy of the prediction result is improved.
Drawings
FIG. 1 is a schematic flow chart of the AI hemodialysis protocol generation method of the invention;
FIG. 2 is a schematic diagram of a model for generating an AI hemodialysis protocol according to the present invention;
FIG. 3A is a top half of a detailed flow chart of the AI hemodialysis protocol generation method of the present invention;
FIG. 3B is a lower half of a detailed flow chart of the AI hemodialysis protocol generation method of the present invention;
FIG. 4 is a ROC plot of inclusion versus non-inclusion for dialysis time;
FIG. 5 is a ROC plot of inclusion versus non-inclusion for a weekly dose of EPO;
FIG. 6 is a ROC plot of inclusion versus non-inclusion for dialysate K concentration;
figure 7 is a ROC plot for inclusion versus non-inclusion for dialysate Ca concentration;
fig. 8 is a ROC plot for inclusion versus non-inclusion for dialysate Na concentration;
FIG. 9 is a ROC plot for dialysate flow with and without inclusion;
FIG. 10 is a ROC plot of incorporation versus non-incorporation for conductance;
FIG. 11 is a ROC plot of inclusion versus non-inclusion for dialysate temperature;
FIG. 12 non-numeric data transformation rules;
FIG. 13 is a graph of whether the conclusion of the dialyzer model is applied as an input variable to the significance test of the dialysis setup model;
FIG. 14 is a graph of whether the conclusion of the dialysis setup model is applied to the dialysis agent model significance test as an input variable;
FIG. 15 is a graph of whether the conclusion of the dialysate agent model is applied to the dialysate model significance test as an input variable;
FIG. 16 shows MSE values corresponding to respective indices;
FIG. 17 shows MSE values corresponding to respective indices;
FIG. 18 achievement rates under criteria;
FIG. 19 shows the occurrence rate of abnormal conditions in each group.
Detailed Description
Example 1:
as shown in fig. 2, 3A, 3B, an AI hemodialysis protocol generation model includes a dialyzer submodel, a dialysis setting submodel, a dialysis agent submodel, and a dialysate submodel;
a dialyzer submodel; the dialyzer submodel comprises a dialyzer model component and a dialysis mode component;
the dialyzer model component comprises a dialyzer model algorithm component and a dialyzer model feature importance analysis component; the dialyzer model component gives the dialyzer model component input characteristics through the dialyzer model characteristic importance degree analysis component, inputs the dialyzer model component input characteristics into the dialyzer model algorithm component, and finally gives dialyzer model parameter values through the dialyzer model algorithm component;
further, the dialyzer model feature importance analyzing component performs feature importance analysis on the input features by using an xgboost algorithm, fits all the input features and the actual dialyzer model by using the xgboost, ranks the average reduction of the loss function when each input feature is used as a partition attribute from large to small to obtain a feature importance ranking, and uses the input feature ranked at the top as the dialyzer model algorithm component input feature.
Preferably, the input features of the top fifteen are used as the input features of the dialyzer model algorithm component.
The dialyzer model algorithm component is composed of a decision classification tree, the decision classification tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold of the input feature, the input features are classified layer by layer according to the thresholds, and finally the input features are classified into a certain dialyzer category;
the dialysis mode component comprises a dialysis mode algorithm component and a dialysis mode feature importance analysis component; the dialysis mode component firstly gives the input characteristics of the dialysis mode component through the dialysis mode characteristic importance degree analysis component, and then inputs the input characteristics of the dialysis mode component into the dialysis mode algorithm component; finally, the dialysis mode algorithm component provides dialysis mode parameter values.
Furthermore, the dialysis mode feature importance analysis module performs feature importance analysis on the input features by using an xgboost algorithm, fits all the input features and the dialysis modes by using an xgboost, sorts the average reduction amount of the loss function when each input feature is used as a partition attribute from large to small to obtain a feature importance ranking, and uses the input feature ranked at the top as the input feature of the dialysis mode algorithm module.
Preferably, the input features of the top fifteen are used as the input features of the dialysis mode algorithm component.
The dialysis mode algorithm component is composed of a logistic regression algorithm, and firstly, input features are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysis mode; the dialysis mode with the maximum probability is the output result of the dialysis mode algorithm component;
the dialyzer model parameter values and dialysis mode parameter values given by the dialyzer submodel constitute the most basic content in the fluid dialysis protocol.
A dialysis setting sub-model; the dialysis setting submodel comprises a preset dehydration component, a dialysis time component and a dialysis blood flow component;
the preset dehydration component comprises a preset dehydration algorithm component and a preset dehydration characteristic importance degree analysis component; the preset dehydration component firstly gives the input characteristics of the preset dehydration component through the preset dehydration characteristic importance analysis component; then, inputting the input characteristics of the preset dehydration component into a preset dehydration algorithm component; finally, a preset dehydration algorithm component gives a preset dehydration value parameter;
further, the preset dehydration characteristic importance degree analysis component analyzes the input characteristics by using an xgboost algorithm, fits all the input characteristics and the actual preset dehydration value by using the xgboost, sorts the average reduction quantity of the loss function when each characteristic is used as a partition attribute from large to small to obtain characteristic importance degree sorting, and takes the input characteristic with the top ranking as the input characteristic of the preset dehydration algorithm component;
preferably, the input features of the top fifteen are taken as the input features of the preset dehydration algorithm component.
The preset dehydration algorithm component is composed of a BP neural network regression algorithm, the BP neural network regression algorithm is composed of a plurality of layers of hidden layers and an output layer, each layer of hidden layers and the output layer are composed of a plurality of neurons, each neuron in the first hidden layer combines input features in a multiple regression equation mode, an equation result is subjected to nonlinear transformation through an activation function, then the neuron of each hidden layer performs multiple regression on data obtained by the previous layer, the equation result is subjected to nonlinear transformation through the activation function, the number of the neurons of the output layer is 1, only previous numerical values are subjected to multiple linear regression, and the final value output by the multiple linear regression is the preset dehydration value predicted by the model.
The dialysis time component comprises a dialysis time algorithm component and a dialysis time characteristic importance degree analysis component; the dialysis time component firstly gives the input characteristics of the dialysis time component through the dialysis time characteristic importance degree analysis component; then, inputting the input characteristics of the dialysis time component into a dialysis time algorithm component; finally, a dialysis time algorithm component gives the dialysis time parameter.
Further, the dialysis time feature importance degree analysis module performs feature importance degree analysis on the input features by using an xgboost algorithm, fits all the input features and the dialysis time values by using the xgboost, sorts the average reduction amount of the loss function of each feature when the feature is used as a partition attribute from large to small to obtain a feature importance degree sort, and uses the input feature with the top rank as the input feature of the dialysis time algorithm module;
preferably, the input features of the top fifteen are used as the input features of the dialysis time algorithm component.
The dialysis time algorithm component is composed of classified random forests, each random forest is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the random forest is the dialysis time parameter.
The dialysis blood flow component comprises a dialysis blood flow algorithm component and a dialysis blood flow characteristic importance analysis component; firstly, the dialysis blood flow algorithm component gives the input characteristics of the dialysis blood flow component through a dialysis blood flow characteristic importance degree analysis component; then, inputting the input characteristics of the dialysis blood flow volume component into a dialysis blood flow volume algorithm component; finally, a dialysis blood flow algorithm component gives a dialysis blood flow parameter.
Furthermore, the dialysis blood flow characteristic importance degree analysis component performs characteristic importance degree analysis on the input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysis blood flow value by using the xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristics are used as the partition attributes from large to small to obtain a characteristic importance degree ranking, and uses the input characteristic ranked at the top as the input characteristic of the dialysis blood flow algorithm component.
Preferably, the input features of the top fifteen are used as the input features of the dialysis blood flow algorithm component.
The blood flow algorithm component is composed of a multiple linear regression equation, and the input characteristics are substituted into the multiple linear regression equation to obtain the final numerical value, namely the value of the blood flow parameter.
A dialysis agent sub-model; the dialysis agent sub-model comprises an anticoagulant component and an EPO weekly dose component;
the anticoagulant component comprises an anticoagulant algorithm component and an anticoagulant characteristic importance analysis component;
the anticoagulant component firstly gives the input characteristics of the anticoagulant component through the anticoagulant characteristic importance analysis component; then inputting the anticoagulant component input characteristics into an anticoagulant algorithm component; finally, the anticoagulant algorithm component gives anticoagulant dosage parameters;
furthermore, the anticoagulant characteristic importance degree analysis module analyzes the characteristic importance degree of the input characteristics by using an xgboost algorithm, fits all the input characteristics and the anticoagulant dosage by using xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristics are used as the partition attributes from large to small to obtain characteristic importance degree sorting, and uses the input characteristics with the top ranking as the input characteristics of the anticoagulant algorithm module.
Preferably, the input features of the top fifteen are used as the input features of the anticoagulant algorithm component.
The anticoagulant dose algorithm component is composed of a decision regression tree algorithm, the decision regression tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold value of the input feature, the nodes are divided layer by layer according to the threshold values, the final output result is a mean value of numerical values in the node, and the mean value of the numerical values in the node is the anticoagulant dose.
The EPO week dosage component comprises an EPO week dosage algorithm component and an EPO week dosage characteristic importance analysis component; the EPO week dosage component firstly gives EPO week dosage component input characteristics through an EPO week dosage characteristic importance analysis component; then, the EPO weekly dose component input features are input into the EPO weekly dose algorithm component; finally, the EPO weekly dose algorithm component gives the EPO weekly dose parameter;
furthermore, the EPO week dose feature importance degree analysis component analyzes the feature importance degree of the input features by using an xgboost algorithm, fits all the input features and EPO week dose by using an xgboost, sorts the average reduction quantity of the loss function of each feature when the features are used as the division attributes from large to small to obtain a feature importance degree sort, and uses the input features with the top rank as the input features of the EPO week dose algorithm component.
Preferably, the input features of the top fifteen are used as input features of the EPO weekly dosage algorithm component.
Further, the EPO refers to a drug intended for use in a dialysis regimen; the EPO weekly dosage refers to the dosage of one week of medication in a dialysis scheme.
The EPO weekly dose algorithm component is composed of a decision regression tree, the decision regression tree is composed of a plurality of nodes, each node is provided with an input characteristic and a threshold of the input characteristic, the nodes are divided layer by layer according to the threshold, the final output result is the mean value of numerical values in the nodes, and the mean value of the numerical values in the nodes is the EPO weekly dose.
A dialysate sub-model; the dialysate submodel comprises a dialysate flow component, a dialysate temperature component, a dialysate conductivity component, a dialysate potassium concentration component, a dialysate sodium concentration component and a dialysate calcium concentration component;
the dialysate flow component comprises a dialysate flow algorithm component and a dialysate flow characteristic importance analysis component; firstly, the dialysate flow component gives input characteristics of the dialysate flow component through a dialysate flow characteristic importance analysis component; then inputting the dialysate flow component input characteristics into a dialysate flow algorithm component; finally, the dialysate flow algorithm component gives dialysate flow parameters;
furthermore, the dialysate flow characteristic importance analysis module performs characteristic importance analysis on the input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate flow by using an xgboost, sorts the average reduction of loss functions of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance ranking, and uses the input characteristic ranked at the top as the input characteristic of the dialysate flow algorithm module.
Preferably, the top fifteen input features are used as dialysate flow algorithm component input features.
The dialysate flow algorithm component is composed of a logistic regression algorithm, and firstly, input characteristics are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysate flow; the dialysate flow with the maximum probability is the output result of the dialysate flow algorithm component;
the dialysate temperature component comprises a dialysate temperature algorithm component and a dialysate temperature characteristic importance analysis component; the dialysate temperature component firstly gives input characteristics of the dialysate temperature component through a dialysate temperature characteristic importance analysis component; then inputting the input characteristics of the dialysate temperature component into a dialysate temperature algorithm component; finally, the dialysate temperature algorithm component gives a dialysate temperature parameter;
further, the dialysate temperature feature importance analyzing component uses an xgboost algorithm to analyze the feature importance of the input features, fits all the input features and the dialysate temperature using an xgboost, sorts the average reduction of the loss function of each feature when the feature is used as a partition attribute from large to small to obtain a feature importance ranking, and uses the input feature with the top ranking as the input feature of the dialysate temperature algorithm component;
preferably, the input features of the top fifteen are used as the dialysate temperature algorithm component input features.
The dialysate temperature algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate temperature parameter.
The dialysate conductivity component comprises a dialysate conductivity algorithm component and a dialysate conductivity feature importance analysis component; the dialysate conductivity component firstly gives input characteristics of the dialysate conductivity component through the dialysate conductivity characteristic importance degree analysis component; then inputting the input characteristics of the dialysate conductivity component into a dialysate conductivity algorithm component; finally, the dialysate conductivity algorithm component gives parameters of the conductivity of the dialysate;
furthermore, the dialysate conductivity feature importance analysis module analyzes the feature importance of the input features by using an xgboost algorithm, fits all the input features and the dialysate conductivity by using an xgboost, sorts the average reduction of loss functions of each feature as a division attribute from large to small to obtain a feature importance ranking, and takes the input feature ranked at the top as the input feature of the dialysate conductivity algorithm module.
Preferably, the input features of the top fifteen are used as the dialysate conductivity algorithm component input features.
The dialysate conductivity algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely, the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the conductivity parameter of the dialysate.
The dialysate potassium concentration component comprises a dialysate potassium concentration algorithm component and a dialysate potassium concentration characteristic importance analysis component; firstly, the dialysate potassium concentration component gives input characteristics of the dialysate potassium concentration component through a dialysate potassium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate potassium concentration component into a dialysate potassium concentration algorithm component; finally, a dialysate potassium concentration algorithm component gives a dialysate potassium concentration parameter;
furthermore, the dialysate potassium concentration feature importance analysis module performs feature importance analysis on the input features by using an xgboost algorithm, fits all the input features and the dialysate potassium concentration by using an xgboost, sorts the average reduction amount of the loss function of each feature as a division attribute from large to small to obtain a feature importance ranking, and takes the input feature ranked at the top as the input feature of the dialysate potassium concentration algorithm module.
Preferably, the input features of the top fifteen are used as the dialysate potassium concentration algorithm component input features.
The dialysate potassium concentration algorithm component is composed of a classified random forest which is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate potassium concentration parameter.
The dialysate sodium concentration component comprises a dialysate sodium concentration algorithm component and a dialysate sodium concentration characteristic importance analysis component; firstly, the input characteristics of the dialysate sodium concentration component are given by the dialysate sodium concentration characteristic importance analysis component; then inputting the input characteristics of the dialysate sodium concentration component into a dialysate sodium concentration algorithm component; finally, the dialysate sodium concentration algorithm component gives parameters of dialysate sodium concentration;
furthermore, the dialysate sodium concentration characteristic importance analysis module performs characteristic importance analysis on the input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate sodium concentration by using an xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristic is used as a division attribute from large to small to obtain characteristic importance sorting, and uses the input characteristic with the top ranking as the input characteristic of the dialysate sodium concentration algorithm module.
Preferably, the input features of the top fifteen are used as the input features of the dialysate sodium concentration algorithm component.
The dialysate sodium concentration algorithm component is composed of a BP neural network classification algorithm; the other structures of the BP neural network classification algorithm except an output layer are the same as the BP neural network regression algorithm, the output layer of the BP neural network classification algorithm is the classification category number of dialysate sodium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of the hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to the range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate sodium concentration category is obtained.
The dialysate calcium concentration component comprises a dialysate calcium concentration algorithm component and a dialysate calcium concentration characteristic importance analysis component; firstly, the dialysate calcium concentration component gives input characteristics of the dialysate calcium concentration component through a dialysate calcium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate calcium concentration component into a dialysate calcium concentration algorithm component; finally, the dialysate calcium concentration algorithm component provides a dialysate calcium concentration parameter.
Furthermore, the dialysate calcium concentration characteristic importance degree analysis module performs characteristic importance degree analysis on the input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate calcium concentration by using an xgboost, sorts the average reduction amount of the loss function of each characteristic from large to small when the characteristics are used as the partition attributes to obtain a characteristic importance degree sequence, and uses the input characteristic ranked at the top as the input characteristic of the dialysate calcium concentration algorithm module.
Preferably, the input features of the top fifteen are used as the dialysate calcium concentration algorithm component input features.
The dialysate calcium concentration algorithm component is composed of a BP neural network classification algorithm; the BP neural network classification algorithm is the same as a BP neural network regression algorithm in structure except for an output layer, the output layer of the BP neural network classification algorithm is the number of classification categories of dialysate calcium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of a hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to a range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate calcium concentration category is obtained.
Finally, the AI hemodialysis plan generating model combines the dialysis mode parameter value, the dialyzer model parameter value, the pre-dehydration value parameter, the dialysis time parameter, the dialysis blood flow parameter, the anticoagulant dosage parameter, the EPO weekly dosage parameter, the dialysate flow parameter, the dialysate temperature parameter, the dialysate conductivity parameter, the dialysate potassium concentration parameter, the dialysate sodium concentration parameter and the dialysate calcium concentration parameter to generate the hemodialysis plan.
The AI hemodialysis scheme generation model can assist medical personnel to formulate an individualized dialysis scheme, and reduce the labor intensity and professional requirements of the medical personnel. Secondly, can reduce the dialysis risk, reduce or avoid the risk that takes place the adverse reaction in the dialysis, guarantee patient's safety, improve medical care quality. Thirdly, using the conclusion of the dialyzer submodel as an input variable to be applied to a dialysis setting submodel; using the conclusion of the dialysis setting submodel as an input variable to be applied to the dialysis medicament submodel; using the conclusion of the dialysis agent submodel as an input variable to be applied to the dialysate submodel; the method can reduce the prediction errors of the pre-dehydration value parameter, the dialysis time parameter, the dialysis blood flow parameter, the anticoagulant dosage parameter, the EPO week dosage parameter, the dialysate flow parameter, the dialysate temperature parameter, the dialysate conductivity parameter, the dialysate potassium concentration parameter, the dialysate sodium concentration parameter and the dialysate calcium concentration parameter, and improve the accuracy of the prediction result.
According to the method, the input features which are ranked at the top are used as the input features of all the components, particularly the input features which are ranked at the top fifteen are used as the input features of all the components, on one hand, the inaccuracy of a prediction result caused by incomplete medical feature data is avoided, on the other hand, important medical features can be highlighted, and the accuracy of the prediction result is improved.
Example 2: as shown in fig. 1 and 3, on the basis of embodiment 1, a method for generating a hemodialysis protocol based on an AI hemodialysis protocol generation model, which includes a dialyzer submodel, a dialysis setting submodel, a dialysis agent submodel and a dialysate submodel, four submodels; the method comprises the following steps:
s1, inputting the primary input characteristics into a dialyzer submodel, wherein the dialyzer submodel provides dialysis mode parameter values and dialyzer model parameter values;
the primary input features are derived from a hemodialysis database of a medical institution, and comprise basic information, pre-dialysis grid examination, historical dialysis schemes and historical examination and examination data;
a dialyzer submodel; the dialyzer submodel comprises a dialyzer model component and a dialysis mode component;
the dialyzer model component comprises a dialyzer model algorithm component and a dialyzer model feature importance analysis component;
the dialyzer model component gives dialyzer model component input characteristics through the dialyzer model characteristic importance analysis component, inputs the dialyzer model component input characteristics into the dialyzer model algorithm component, and finally gives dialyzer model parameter values through the dialyzer model algorithm component;
further, the dialyzer model feature importance analyzing component performs feature importance analysis on the primary input features by using an xgboost algorithm, fits all the primary input features with the actual dialyzer model by using the xgboost, ranks the average reduction of the loss function when each input feature of the primary input features is used as a partition attribute from large to small to obtain a feature importance ranking, and uses the input feature ranked at the top as the dialyzer model algorithm component input feature.
The dialyzer model algorithm component is composed of a decision classification tree, the decision classification tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold of the input feature, the input features are classified layer by layer according to the thresholds, and finally the input features are classified into a certain dialyzer category; finally, the dialyzer model algorithm component gives dialyzer model parameter values;
the dialysis mode component comprises a dialysis mode algorithm component and a dialysis mode feature importance analysis component; the dialysis mode component firstly gives the input characteristics of the dialysis mode component through the dialysis mode characteristic importance degree analysis component, and then inputs the input characteristics of the dialysis mode component into the dialysis mode algorithm component; finally, the dialysis mode algorithm component provides dialysis mode parameter values.
Furthermore, the dialysis mode feature importance degree analysis module performs feature importance degree analysis on the primary input features by using an xgboost algorithm, fits all the input features and the dialysis modes by using the xgboost, ranks the average reduction amount of the loss function when each input feature is used as a partition attribute from large to small to obtain a feature importance degree rank, and uses the input feature ranked at the top as the input feature of the dialysis mode algorithm module.
The dialysis mode algorithm component is composed of a logistic regression algorithm, and firstly, input features are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysis mode; the dialysis mode with the maximum probability is the output result of the dialysis mode algorithm component;
the dialyzer model parameter values and dialysis mode parameter values given by the dialyzer submodel constitute the most basic content in the fluid dialysis protocol.
S2, combining the primary input characteristics with dialysis mode parameter values and dialyzer model parameter values to generate secondary input characteristics;
s3, inputting the secondary input characteristics into a dialysis setting sub-model, and outputting a pre-dehydration value parameter, a dialysis time parameter and a dialysis blood flow parameter by the dialysis setting sub-model;
a dialysis setting sub-model; the dialysis setting submodel comprises a preset dehydration component, a dialysis time component and a dialysis blood flow component;
the preset dehydration component comprises a preset dehydration algorithm component and a preset dehydration characteristic importance degree analysis component; the preset dehydration component firstly gives the input characteristics of the preset dehydration component through the preset dehydration characteristic importance analysis component; then, inputting the input characteristics of the preset dehydration component into a preset dehydration algorithm component; finally, the preset dehydration algorithm component gives a preset dehydration value parameter;
further, the preset dehydration characteristic importance degree analysis component analyzes the secondary input characteristics by using an xgboost algorithm, fits all the secondary input characteristics and the actual preset dehydration value by using the xgboost, sorts the average reduction amount of the loss function when each characteristic is used as the partition attribute from large to small to obtain a characteristic importance degree sorting, and uses the input characteristic with the top ranking as the input characteristic of the preset dehydration algorithm component;
the preset dehydration algorithm component is composed of a BP neural network regression algorithm, the BP neural network regression algorithm is composed of a plurality of layers of hidden layers and an output layer, each layer of hidden layers and the output layer are composed of a plurality of neurons, each neuron in the first hidden layer combines input features in a multiple regression equation mode, an equation result is subjected to nonlinear transformation through an activation function, then the neuron of each hidden layer performs multiple regression on data obtained by the previous layer, the equation result is subjected to nonlinear transformation through the activation function, the number of the neurons of the output layer is 1, only previous numerical values are subjected to multiple linear regression, and the final value output by the multiple linear regression is the preset dehydration value predicted by the model.
The dialysis time component comprises a dialysis time algorithm component and a dialysis time characteristic importance degree analysis component; the dialysis time component firstly gives the input characteristics of the dialysis time component through the dialysis time characteristic importance degree analysis component; then, inputting the input characteristics of the dialysis time component into a dialysis time algorithm component; finally, a dialysis time algorithm component gives the dialysis time parameter.
Further, the dialysis time feature importance analysis module analyzes the feature importance of the secondary input features by using an xgboost algorithm, fits all the secondary input features and the dialysis time values by using an xgboost, sorts the average reduction amount of the loss function of each feature when the feature is used as a division attribute from large to small to obtain a feature importance ranking, and uses the input feature with the top ranking as the input feature of the dialysis time algorithm module;
the dialysis time algorithm component is composed of classified random forests, each random forest is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the random forest is the dialysis time parameter.
The dialysis blood flow component comprises a dialysis blood flow algorithm component and a dialysis blood flow characteristic importance analysis component; firstly, the dialysis blood flow algorithm component gives the input characteristics of the dialysis blood flow component through a dialysis blood flow characteristic importance degree analysis component; then, inputting the input characteristics of the dialysis blood flow volume component into a dialysis blood flow volume algorithm component; finally, a dialysis blood flow algorithm component gives a dialysis blood flow parameter.
Furthermore, the dialysis blood flow characteristic importance degree analysis module analyzes the characteristic importance degree of the secondary input characteristics by using an xgboost algorithm, fits all the secondary input characteristics and the dialysis blood flow value by using the xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristics are used as the division attributes from large to small to obtain characteristic importance degree sorting, and takes the input characteristic with the top ranking as the input characteristic of the dialysis blood flow algorithm module.
The blood flow algorithm component is composed of a multiple linear regression equation, and the input characteristics are substituted into the multiple linear regression equation to obtain a final numerical value, namely the value of the blood flow parameter.
S4, combining the primary input features with the pre-dehydration value parameters, the dialysis time parameters and the dialysis blood flow parameters to generate tertiary input features;
s5, inputting the three-time input characteristics into a dialysis agent sub-model, and giving anticoagulant dose parameters and EPO (erythropoietin) weekly dose parameters by the dialysis agent sub-model;
the dialysis agent sub-model comprises an anticoagulant component and an EPO weekly dose component;
the anticoagulant component comprises an anticoagulant algorithm component and an anticoagulant characteristic importance analysis component;
the anticoagulant component firstly gives the input characteristics of the anticoagulant component through the anticoagulant characteristic importance analysis component; then inputting the input characteristics of the anticoagulant component into the anticoagulant algorithm component; finally, the anticoagulant algorithm component gives anticoagulant dosage parameters;
furthermore, the anticoagulant characteristic importance degree analysis module analyzes the characteristic importance degree of the three input characteristics by using an xgboost algorithm, fits all the input characteristics and the anticoagulant dosage by using xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristics are used as the partition attributes from large to small to obtain characteristic importance degree sorting, and uses the input characteristic ranked at the top as the input characteristic of the anticoagulant algorithm module.
The anticoagulant dose algorithm component is composed of a decision regression tree algorithm, the decision regression tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold of the input feature, the nodes are divided layer by layer according to the threshold, the final output result is the mean value of numerical values in the nodes, and the mean value of the numerical values in the nodes is the anticoagulant dose.
The EPO week dosage component comprises an EPO week dosage algorithm component and an EPO week dosage characteristic importance analysis component; the EPO week dosage component firstly gives EPO week dosage component input characteristics through an EPO week dosage characteristic importance analysis component; then, the EPO weekly dose component input features are input into the EPO weekly dose algorithm component; finally, the EPO weekly dose algorithm component gives the EPO weekly dose parameter;
furthermore, the EPO week dose characteristic importance degree analysis component analyzes the characteristic importance degree of the three input characteristics by using an xgboost algorithm, all the input characteristics and the EPO week dose are fitted by using the xgboost, the average reduction quantity of the loss function of each characteristic when the characteristic is used as the dividing attribute is sorted from large to small to obtain the characteristic importance degree sort, and the input characteristic with the top rank is used as the input characteristic of the EPO week dose algorithm component.
Further, the EPO refers to a drug intended for use in a dialysis regimen; the EPO weekly dose refers to the dose of a planned week of administration in a dialysis regimen.
The EPO weekly dose algorithm component is composed of a decision regression tree, the decision regression tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold of the input feature, the nodes are divided layer by layer according to the thresholds, the final output result is the mean value of numerical values in the nodes, and the mean value of the numerical values in the nodes is the EPO weekly dose.
S6, combining the primary input characteristics with anticoagulant dose parameters and EPO week dose parameters to generate four-time input characteristics;
s7, inputting the four input characteristics into a dialysate submodel, wherein the dialysate submodel gives a dialysate flow parameter, a dialysate temperature parameter, a dialysate conductivity parameter, a dialysate potassium concentration parameter, a dialysate sodium concentration parameter and a dialysate calcium concentration parameter;
the dialysate submodel comprises a dialysate flow component, a dialysate temperature component, a dialysate conductivity component, a dialysate potassium concentration component, a dialysate sodium concentration component and a dialysate calcium concentration component;
the dialysate flow component comprises a dialysate flow algorithm component and a dialysate flow characteristic importance analysis component; firstly, the dialysate flow component gives input characteristics of the dialysate flow component through a dialysate flow characteristic importance analysis component; then inputting the dialysate flow component input characteristics into a dialysate flow algorithm component; finally, the dialysate flow algorithm component gives dialysate flow parameters;
furthermore, the dialysate flow characteristic importance analysis module analyzes the characteristic importance of the three input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate flow by using an xgboost, sorts the average reduction of loss functions of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance ranking, and uses the input characteristic with the top ranking as the input characteristic of the dialysate flow algorithm module.
The dialysate flow algorithm component is composed of a logistic regression algorithm, and firstly, input characteristics are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysate flow; the dialysate flow with the maximum probability is the output result of the dialysate flow algorithm component;
the dialysate temperature component comprises a dialysate temperature algorithm component and a dialysate temperature feature importance analysis component; the dialysate temperature component firstly gives input characteristics of the dialysate temperature component through a dialysate temperature characteristic importance analysis component; then inputting the dialysate temperature component input characteristics into a dialysate temperature algorithm component; finally, the dialysate temperature algorithm component gives a dialysate temperature parameter;
further, the dialysate temperature feature importance analyzing component uses an xgboost algorithm to analyze feature importance of the three input features, fits all the input features and the dialysate temperature using the xgboost, sorts the average reduction of loss functions of each feature as a partition attribute from large to small to obtain a feature importance ranking, and uses the input feature with the top ranking as the input feature of the dialysate temperature algorithm component;
the dialysate temperature algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate temperature parameter.
The dialysate conductivity component comprises a dialysate conductivity algorithm component and a dialysate conductivity feature importance analysis component; the dialysate conductivity component firstly gives input characteristics of the dialysate conductivity component through the dialysate conductivity characteristic importance degree analysis component; then, inputting the input characteristics of the dialysate conductivity component into a dialysate conductivity algorithm component; finally, the dialysate conductivity algorithm component gives parameters of the conductivity of the dialysate;
furthermore, the dialysate conductivity feature importance analysis module performs feature importance analysis on the three input features by using an xgboost algorithm, fits all the input features and the dialysate conductivity by using the xgboost, sorts the average reduction of the loss function of each feature as a partition attribute from large to small to obtain a feature importance ranking, and takes the input feature ranked at the top as the input feature of the dialysate conductivity algorithm module.
The dialysate conductivity algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the conductivity parameter of the dialysate.
The dialysate potassium concentration component comprises a dialysate potassium concentration algorithm component and a dialysate potassium concentration characteristic importance analysis component; firstly, the input characteristics of the dialysate potassium concentration component are given by the dialysate potassium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate potassium concentration component into a dialysate potassium concentration algorithm component; finally, a dialysate potassium concentration algorithm component gives a dialysate potassium concentration parameter;
furthermore, the dialysate potassium concentration characteristic importance degree analysis module performs characteristic importance degree analysis on the three input characteristics by using an xgboost algorithm, all the input characteristics and the dialysate potassium concentration are fitted by using the xgboost algorithm, the average reduction amount of the loss function of each characteristic when the characteristics are used as the partition attributes is sorted from large to small to obtain characteristic importance degree sorting, and the input characteristic which is ranked at the top is used as the input characteristic of the dialysate potassium concentration algorithm module.
The dialysate potassium concentration algorithm component is composed of a classified random forest which is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate potassium concentration parameter.
The dialysate sodium concentration component comprises a dialysate sodium concentration algorithm component and a dialysate sodium concentration characteristic importance analysis component; firstly, the dialysate sodium concentration component gives input characteristics of the dialysate sodium concentration component through a dialysate sodium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate sodium concentration component into a dialysate sodium concentration algorithm component; finally, the dialysate sodium concentration algorithm component gives parameters of dialysate sodium concentration;
furthermore, the dialysate sodium concentration characteristic importance degree analysis module performs characteristic importance degree analysis on the three input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate sodium concentration by using the xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance degree sort, and uses the input characteristic ranked at the top as the input characteristic of the dialysate sodium concentration algorithm module.
The dialysate sodium concentration algorithm component is composed of a BP neural network classification algorithm; the BP neural network classification algorithm is the same as a BP neural network regression algorithm in structure except for an output layer, the output layer of the BP neural network classification algorithm is the number of classification categories of dialysate sodium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of a hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to a range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate sodium concentration category is obtained.
The dialysate calcium concentration component comprises a dialysate calcium concentration algorithm component and a dialysate calcium concentration characteristic importance analysis component; firstly, the dialysate calcium concentration component gives the input characteristics of the dialysate calcium concentration component through a dialysate calcium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate calcium concentration component into a dialysate calcium concentration algorithm component; finally, the dialysate calcium concentration algorithm component provides a dialysate calcium concentration parameter.
Furthermore, the dialysate calcium concentration characteristic importance degree analysis module performs characteristic importance degree analysis on the three input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate calcium concentration by using the xgboost, sorts the average reduction quantity of the loss function of each characteristic from large to small when the characteristics are used as the partition attributes to obtain a characteristic importance degree sequence, and uses the input characteristic ranked at the top as the input characteristic of the dialysate calcium concentration algorithm module.
The dialysate calcium concentration algorithm component is composed of a BP neural network classification algorithm; the BP neural network classification algorithm is the same as a BP neural network regression algorithm in structure except for an output layer, the output layer of the BP neural network classification algorithm is the number of classification categories of dialysate calcium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of a hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to a range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate calcium concentration category is obtained.
And S8, combining the dialysis mode parameter value, the dialyzer model parameter value, the pre-dehydration value parameter, the dialysis time parameter, the dialysis blood flow parameter, the anticoagulant dosage parameter, the EPO week dosage parameter, the dialysate flow parameter, the dialysate temperature parameter, the dialysate conductivity parameter, the dialysate potassium concentration parameter, the dialysate sodium concentration parameter and the dialysate calcium concentration parameter to generate a hemodialysis scheme.
Further, the basic information data comprises age, sex, height, weight, blood type, nationality, dialysis age, primary morbidity, infection history and important medical history;
the pre-dialysis lattice examination comprises dry body weight, last off-machine body weight, pre-dialysis body weight, blood pressure (low pressure), blood pressure (high pressure), heart rate and body temperature;
the historical dialysis scheme comprises 13 indexes including a dialysis mode, a dialyzer, preset dehydration, dialysis time, anticoagulant dose, blood flow, EPO weekly dose, dialysate K concentration, dialysate Ca concentration, dialysate Na concentration, dialysate flow, conductivity and dialysate temperature;
the historical test examination data includes hemoglobin, mean hemoglobin concentration, mean hemoglobin content, platelet count, white blood cell count, red blood cell specific volume, mean red blood cell volume, creatinine, urea, uric acid, albumin, total protein, prealbumin, glucose, sodium, potassium, calcium, phosphorus, chlorine, bicarbonate, plasma fibrinogen, fibrin (ogen) degradation products, plasma D-D dimer, serum troponin I, serum troponin T, myoglobin, serum iron, serum total iron binding capacity, serum ferritin, parathyroid hormone, occult blood, urine protein.
Further, step S1 further comprises a sub-step of giving an input feature, comprising:
s11, collecting characteristic data;
acquiring existing hemodialysis overall process data in a hemodialysis database of a medical institution, and selecting dialysis data of a maintenance dialysis patient, wherein the dialysis data comprises basic information data, dialysis precursor grid inspection data, historical dialysis scheme data and historical inspection data;
the basic information data comprises age, sex, height, weight, blood type, nationality, dialysis age, primary morbidity, infection history and important medical history;
the pre-dialysis check data comprises dry body weight, last off-machine body weight, pre-dialysis body weight, blood pressure (low pressure), blood pressure (high pressure), heart rate and body temperature;
the historical dialysis scheme data comprises 13 indexes including a dialysis mode, a dialyzer, preset dehydration, dialysis time, anticoagulant dose, blood flow, EPO weekly dose, dialysate K concentration, dialysate Ca concentration, dialysate Na concentration, dialysate flow, conductivity and dialysate temperature;
the historical test examination data includes hemoglobin, mean hemoglobin concentration, mean hemoglobin content, platelet count, white blood cell count, red blood cell specific volume, mean red blood cell volume, creatinine, urea, uric acid, albumin, total protein, prealbumin, glucose, sodium, potassium, calcium, phosphorus, chloride, bicarbonate, plasma fibrinogen, fibrin (ogen) degradation products, plasma D-D dimer, serum troponin I, serum troponin T, myoglobin, serum iron, serum total iron binding capacity, serum ferritin, parathyroid hormone, occult blood, urine protein.
S12, cleaning the characteristic data to obtain standardized characteristic data; the characteristic data comprises numerical characteristic data and non-numerical characteristic data;
s121, performing unit conversion on each type of numerical characteristic data, and converting each type of numerical characteristic data into numerical characteristic data of a unified unit in a unified unit;
s122, processing abnormal values in numerical characteristic data of the unified unit;
the abnormal value processing method comprises the following steps:
s1221, if the numerical data in the numerical characteristic data of the unified unit obeys normal distribution, the data in the numerical characteristic data and the data except that the average value of the data in the numerical characteristic data is more than or less than 3 times of the standard deviation are abnormal values; then, eliminating the data of the row where the abnormal value is located; the average value is the average value of the columns where the numerical characteristic data are located;
s1222, if the numerical characteristics in the numerical characteristic data of the unified unit do not obey normal distribution, adopting a box-line graph method; the numerical characteristic data of the unified unit is smaller than Q1-1.5QR, or the numerical characteristic data of the unified unit is larger than Q3+1.5QR to be an abnormal value. Then the data of the row where the abnormal value is located is culled.
Further, Q1 is the lower quartile, Q3 is the upper quartile, and QR is the subtraction of the lower quartile from the upper quartile.
Obtaining numerical characteristic data of a unified unit with abnormal values removed;
s123, synchronizing, and assigning the non-numerical characteristic data by referring to a non-numerical data conversion rule to obtain assigned non-numerical characteristic data;
example (c): the non-numerical data conversion rule is shown in FIG. 12
And S124, finally, forming standardized feature data by the numerical feature data of the unified unit with the abnormal values removed and the assigned non-numerical feature data.
The normalized feature data is a primary input feature.
The method for generating the hemodialysis scheme by the AI hemodialysis scheme generation model can assist medical personnel to formulate an individualized dialysis scheme, and reduce the labor intensity and professional requirements of the medical personnel. Secondly, can reduce the dialysis risk, reduce or avoid the risk that takes place the adverse reaction in the dialysis, guarantee patient's safety, improve medical care quality. Thirdly, the AI hemodialysis scheme generation model is used for generating the hemodialysis scheme, and the conclusion of the dialyzer submodel is used as an input variable for the dialysis setting submodel; using the conclusion of the dialysis setting submodel as an input variable to be applied to the dialysis medicament submodel; using the conclusion of the dialysis agent submodel as an input variable to be applied to the dialysate submodel; the method can reduce the prediction errors of the pre-dehydration value parameter, the dialysis time parameter, the dialysis blood flow parameter, the anticoagulant dosage parameter, the EPO week dosage parameter, the dialysate flow parameter, the dialysate temperature parameter, the dialysate conductivity parameter, the dialysate potassium concentration parameter, the dialysate sodium concentration parameter and the dialysate calcium concentration parameter, and improve the accuracy of the prediction result.
Embodiment 3 the conclusion of the dialyzer submodel is used as an input variable for the dialysis setting submodel; taking the conclusion of the dialysis setting submodel as an input variable, and applying the input variable to a dialysis medicament submodel; using the conclusion of the dialysis agent submodel as an input variable to be applied to the dialysate submodel; the method can reduce the prediction errors of the pre-dehydration value parameter, the dialysis time parameter, the dialysis blood flow parameter, the anticoagulant dosage parameter, the EPO week dosage parameter, the dialysate flow parameter, the dialysate temperature parameter, the dialysate conductivity parameter, the dialysate potassium concentration parameter, the dialysate sodium concentration parameter and the dialysate calcium concentration parameter, and improve the accuracy of the prediction result.
1.1 from the perspective of correlation analysis
As shown in fig. 13, the conclusion of the dialyzer model is applied to the dialysis setting model as an input variable. The dialyzer model is concluded whether there are significant differences as input variables to the results in the dialysis setting model. The P values for the significance test were all less than 0.05, indicating whether there was a significant difference as an input variable to the results in the dialysis setting model. Further, it is described whether the results are significantly correlated with each other in the dialysis setting model as the input variables.
Similarly, as shown in fig. 14, the conclusion of the dialysis setting model is applied to the dialysis drug model as an input variable. And (4) conclusion: whether or not the results in the dialysis agent model are significantly correlated as input variables.
Similarly, as shown in fig. 15, the conclusion of the dialysate agent model is applied to the dialysate model as an input variable. And (4) conclusion: whether the results in the dialysate model are significantly correlated as input variables.
1.2 from the model results
Quantitative data (preset dehydration, blood flow, anticoagulant dose) using MSE as model evaluation criteria; qualitative data (dialysis time, EPO weekly dose, dialysate K concentration, dialysate Ca concentration, dialysate Na concentration, dialysate flow rate, conductivity, dialysate temperature) were evaluated using ROC graph and AUC values as model evaluation criteria.
1.2.1 As shown in FIG. 16, it is assumed whether the conclusion of the dialyzer model is applied to the dialysis setup model as an input variable.
And (4) conclusion: as can be seen from fig. 16 and 4, for the preset dehydration and blood flow in the dialysis setting model results, the corresponding MSE values are both smaller than those for non-inclusion when the conclusion of the dialyzer model is included as input variables. After the prediction is brought into the preset dehydration state, the prediction error of the blood flow is smaller, and the prediction result is more accurate. For the dialysis time, when the conclusion of the dialyzer model is taken as an input variable, the corresponding AUC value (0.86) is greater than the unincorporated AUC value (0.78), which shows that the prediction accuracy of the dialysis time is improved after the incorporation, and the prediction result is more accurate.
To summarize: the conclusion of the dialyzer model is used as an input variable and applied to the dialysis setting model, so that the model prediction result is more accurate.
1.2.2 As shown in FIG. 17, the conclusion of whether or not the dialysis setting model is applied to the dialysate model is used as an input variable.
And (4) conclusion: as can be seen from fig. 17 and 5, for the anticoagulant doses in the dialysate agent model results, when the conclusion of the dialysis setting model is included as an input variable, the corresponding MSE value is less than not included. After the anticoagulant is brought into the blood, the prediction error of the anticoagulant dose is smaller, and the prediction result is more accurate. For the EPO weekly dose, when the conclusion of the inclusion dialysis setting model is taken as an input variable, the corresponding AUC value (0.70) is greater than the non-included AUC value (0.69), which indicates that the prediction accuracy of the EPO weekly dose is improved and the prediction result is more accurate after inclusion.
To summarize: the conclusion of the dialysis setting model is used as an input variable and applied to the dialysis medicament model, so that the model prediction result is more accurate.
1.2.3 whether the conclusion of the dialysate agent model is applied to the dialysate model as an input variable.
And (4) conclusion: as can be seen from fig. 6 to 11, in the dialysate model results, when the conclusion that the dialysate K concentration, the dialysate Ca concentration, the dialysate Na concentration, the dialysate flow rate, the conductivity, and the dialysate temperature were included in the dialysate model is taken as input variables, the corresponding AUC values are all greater than the non-included AUC values, which indicates that the prediction accuracy of the dialysate K concentration, the dialysate Ca concentration, the dialysate Na concentration, the dialysate flow rate, the conductivity, and the dialysate temperature is improved after inclusion, and the prediction result is more accurate.
To summarize: the conclusion of the dialysis agent model is used as an input variable and applied to the dialysate model, so that the model prediction result is more accurate.
Example 4
The clinical verification result conditions of the hemodialysis scheme generated by the AI hemodialysis scheme generation model and the method are as follows:
2.1 percent achievement, as shown in FIG. 18,
with reference to the SOP criterion: blood flow is between 200-400; the dialysis time is between 4 and 4.5 hours; the dialysate flow rate is 500-800; the temperature of the dialysate is 35.5-36.5 ℃; the concentration of the dialysate K is between 0 and 4; the concentration of Ca in the dialysate is between 1.25 and 1.75; the Na concentration of the dialyzate is between 135 and 140.
The doctor's criteria are as follows: aiming at the preset dehydration error, the error is within 0.3; anticoagulant dose error is within 200; the blood flow requirement is in the same interval, wherein the interval I is 200-250, and the interval II is 250-300; the conductivity is required to be in the same grade, wherein the grade I is 13.4-13.7, the grade II is 13.7-14, the grade III is 14-14.3, and the grade IV is 14.3-14.6; other metrics require a perfect match.
And (4) conclusion: under the SOP criterion, the standard-reaching rate exceeds 80%, and under the doctor criterion, the standard-reaching rate exceeds 70%.
2.2 adverse reaction comparison
2.2.1 Adverse reactions can occur during dialysis. The specific contents are as follows:
# adjusting anticoagulant: changing venous kettles, coagulating venous kettles, adjusting anticoagulants, coagulating dialyzer blood, and bleeding tendency # blood pressure fluctuation: modulation of conductance # hypotension: blood filtration pause, ultrafiltration pause, cramp, spasm, sweating, saline fluid infusion, volume-reduced # hypertension (high volume load): adding constant volume, taking the medicine buccally for treating cardiodynia, and treating single super-hypoglycemia: quiet pushing GS, high glucose quiet pushing, hypoglycemia, dysphoria # cardiovascular: oral nitroglycerin, palpitation # other abnormal conditions: body pain, machine temperature, parameter regulation, dizziness, low calcium, emesis and getting off the machine
2.2.2 as shown in fig. 19, validation is performed by means of cross-validation.
Control group: 200 patients are randomly drawn in the current month without using the AI scheme, and the next month with the AI scheme;
experimental groups: 200 patients who differed from the control group were randomly drawn during the month and were not on the AI regimen for the next month.
And (4) conclusion: with the AI scheme, the incidence of each anomaly is generally reduced.
Example 5:
the invention also provides a personalized customized kidney disease hemodialysis scheme system application
The system is characterized in that: the system comprises a user data input module, a dialysis scheme generation module, a customized scheme output module and a dialysis scheme generation module self-learning module;
the user data input module is used for hospital import, batch import or user basic information, pre-dialysis grid examination and historical examination and examination data input. For sending data to a dialysis protocol generation module;
the dialysis scheme generating module is used for generating a dialysis scheme and sending data to the scheme output module;
and the customization scheme output module displays the final personalized hemodialysis customization scheme result through a page.
The hemodialysis scheme generation module is a self-learning module, the model can carry out self-learning iteration, and data generated in operation are added into an original training data set for training again; and after the online operation, continuous iteration is performed according to the data quantity, the dialysis scheme generation model is optimized, and the model accuracy is improved.
Example 6: the invention also provides a system for generating a hemodialysis protocol, which adopts the AI hemodialysis protocol generation model of the invention and/or the method for generating the hemodialysis protocol by the AI hemodialysis protocol generation model of the invention.
Example 7: the invention also provides an application of the hemodialysis program, which is used for assisting medical staff to establish, compare, analyze and verify the hemodialysis program of a patient.
Claims (8)
- An AI hemodialysis protocol generative model modeling method, characterized by: constructing an AI hemodialysis scheme generation model, which comprises a dialyzer submodel, a dialysis setting submodel, a dialysis medicament submodel and a dialysate submodel; using the conclusion of the dialyzer submodel as an input variable for a dialysis setting submodel, using the conclusion of the dialysis setting submodel as an input variable for a dialysis agent submodel, and using the conclusion of the dialysis agent submodel as an input variable for a dialysate submodel;the sub-model of the dialyser gives the parameter value of the model of the dialyser; the dialysis setting sub-model gives a preset dehydration value parameter; the dialysis agent sub-model gives anticoagulant dosage parameters and EPO week dosage parameters; the dialysate submodel gives a dialysate flow parameter, a dialysate temperature parameter, a dialysate conductivity parameter, a dialysate potassium concentration parameter, a dialysate sodium concentration parameter, and a dialysate calcium concentration parameter.
- 2. The AI hemodialysis protocol generative model modeling method of claim 1, wherein:(1) The dialyzer submodel comprises a dialyzer model component and a dialysis mode component;the dialyzer model component comprises a dialyzer model algorithm component and a dialyzer model feature importance analysis component; the dialyzer model feature importance analyzing component gives dialyzer model component input features, and the dialyzer model algorithm component gives dialyzer model parameter values;the dialysis mode component comprises a dialysis mode algorithm component and a dialysis mode feature importance analysis component; the dialysis mode feature importance degree analysis component gives out input features of the dialysis mode component, and the dialysis mode algorithm component gives out values of dialysis mode parameters;(2) The dialysis setting submodel comprises a preset dehydration component, a dialysis time component and a dialysis blood flow component;the preset dehydration component comprises a preset dehydration algorithm component and a preset dehydration characteristic importance degree analysis component; the preset dehydration characteristic importance degree analysis component gives the input characteristics of the preset dehydration component; the preset dehydration algorithm component gives a preset dehydration value parameter;the dialysis time component comprises a dialysis time algorithm component and a dialysis time characteristic importance analysis component; the dialysis time feature importance analysis component gives the input features of the dialysis time component; the dialysis time algorithm component gives a dialysis time parameter;the dialysis blood flow component comprises a dialysis blood flow algorithm component and a dialysis blood flow characteristic importance analysis component; the dialysis blood flow characteristic importance degree analysis component gives an input characteristic of the dialysis blood flow component; the dialysis blood flow algorithm component gives out dialysis blood flow parameters;(3) The dialysis agent sub-model comprises an anticoagulant component and an EPO weekly dose component;the anticoagulant component comprises an anticoagulant algorithm component and an anticoagulant characteristic importance analysis component; the anticoagulant characteristic importance analysis component gives anticoagulant component input characteristics; the anticoagulant algorithm component gives anticoagulant dosage parameters;the EPO week dosage component comprises an EPO week dosage algorithm component and an EPO week dosage characteristic importance analysis component; the EPO week dose characteristic importance analysis component gives EPO week dose component input characteristics; the EPO weekly dose algorithm component gives the EPO weekly dose parameter;(4) The dialysate submodel comprises a dialysate flow component, a dialysate temperature component, a dialysate conductivity component, a dialysate potassium concentration component, a dialysate sodium concentration component and a dialysate calcium concentration component;the dialysate flow component comprises a dialysate flow algorithm component and a dialysate flow characteristic importance analysis component; the dialysate flow characteristic importance analysis component gives input characteristics of the dialysate flow component; the dialysate flow algorithm component gives a dialysate flow parameter;the dialysate temperature component comprises a dialysate temperature algorithm component and a dialysate temperature characteristic importance analysis component; the dialysate temperature characteristic importance analysis module gives input characteristics of the dialysate temperature module; the dialysate temperature algorithm component gives a dialysate temperature parameter;the dialysate conductivity component comprises a dialysate conductivity algorithm component and a dialysate conductivity feature importance analysis component; the dialysate conductivity characteristic importance analysis component provides input characteristics of the dialysate conductivity component; the dialysate conductivity algorithm component gives a dialysate conductivity parameter;the dialysate potassium concentration component comprises a dialysate potassium concentration algorithm component and a dialysate potassium concentration characteristic importance analysis component; the dialysate potassium concentration characteristic importance analysis component gives input characteristics of the dialysate potassium concentration component; a dialysate potassium concentration algorithm component gives a dialysate potassium concentration parameter;the dialysate sodium concentration component comprises a dialysate sodium concentration algorithm component and a dialysate sodium concentration characteristic importance analysis component; the dialysate sodium concentration characteristic importance analysis component gives input characteristics of the dialysate sodium concentration component; the dialysate sodium concentration algorithm component provides parameters of dialysate sodium concentration;the dialysate calcium concentration component comprises a dialysate calcium concentration algorithm component and a dialysate calcium concentration characteristic importance analysis component; the dialysate calcium concentration characteristic importance analysis component gives an input characteristic of the dialysate calcium concentration component; the dialysate calcium concentration algorithm component provides a dialysate calcium concentration parameter.
- 3. The AI hemodialysis protocol generative model modeling method of claim 2, wherein:the dialyzer model feature importance analyzing component performs feature importance analysis on input features by using an xgboost algorithm, fits all the input features and the actual dialyzer model by using the xgboost, ranks the average reduction of the loss function when each input feature is used as a partition attribute from large to small to obtain a feature importance ranking, and takes the input feature ranked at the front as the dialyzer model algorithm component input feature;the dialyzer model algorithm component is composed of a decision classification tree, the decision classification tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold of the input feature, the input features are classified layer by layer according to the thresholds, and finally the input features are classified into a certain dialyzer category;the dialysis mode feature importance degree analysis module performs feature importance degree analysis on input features by using an xgboost algorithm, fits all the input features and the dialysis mode by using the xgboost, ranks the average reduction amount of the loss function when each input feature is used as a partition attribute from large to small to obtain feature importance degree ranks, and takes the input feature ranked at the top as the input feature of the dialysis mode algorithm module;the dialysis mode algorithm component is composed of a logistic regression algorithm, and firstly, input features are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysis mode; the dialysis mode with the maximum probability is the output result of the dialysis mode algorithm component;the preset dehydration characteristic importance degree analysis component analyzes the input characteristics by using an xgboost algorithm, fits all the input characteristics and the actual preset dehydration value by using the xgboost, sorts the average reduction quantity of the loss function when each characteristic is used as a partition attribute from large to small to obtain characteristic importance degree sorting, and takes the input characteristic with the top ranking as the input characteristic of the preset dehydration algorithm component;the preset dehydration algorithm component is composed of a BP neural network regression algorithm, the BP neural network regression algorithm is composed of a plurality of layers of hidden layers and an output layer, each layer of hidden layers and the output layer are composed of a plurality of neurons, each neuron in the first hidden layer combines input features in a multiple regression equation mode, an equation result is subjected to nonlinear transformation through an activation function, then the neuron of each hidden layer performs multiple regression on data obtained by the previous layer, the equation result is subjected to nonlinear transformation through the activation function, the number of the neurons of the output layer is 1, only previous numerical values are subjected to multiple linear regression, and the final value output by the multiple linear regression is the preset dehydration value predicted by the model;the dialysis time characteristic importance analysis module analyzes the characteristic importance of the input characteristic xgboost algorithm, fits all the input characteristics and the dialysis time value by xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristic is used as a division attribute from large to small to obtain characteristic importance sorting, and uses the input characteristic with the top ranking as the input characteristic of the dialysis time algorithm module;the dialysis time algorithm component is composed of classified random forests, each random forest is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the random forest is the dialysis time parameter;the dialysis blood flow characteristic importance degree analysis component analyzes the characteristic importance degree of the input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysis blood flow value by using an xgboost, sorts the average reduction quantity of the loss function when each characteristic is used as a partition attribute from large to small to obtain characteristic importance degree sorting, and takes the input characteristic with the top ranking as the input characteristic of the dialysis blood flow algorithm component;the dialysis blood flow algorithm component is composed of a multiple linear regression equation, and the input characteristics are substituted into the multiple linear regression equation to obtain a final numerical value, namely a value of the blood flow parameter;the anticoagulant feature importance degree analysis component analyzes feature importance degrees of input features by using an xgboost algorithm, all the input features and anticoagulant doses are fitted by using the xgboost, loss function average reduction quantity of each feature when the feature is used as a partition attribute is sorted from large to small to obtain feature importance degree sorting, and the input feature with the top ranking is used as an anticoagulant algorithm component input feature;the anticoagulant dose algorithm component is composed of a decision regression tree algorithm, the decision regression tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold of the input feature, the nodes are divided layer by layer according to the threshold, the final output result is the mean value of numerical values in the nodes, and the mean value of the numerical values in the nodes is the anticoagulant dose;the EPO week dose feature importance degree analysis component analyzes the feature importance degree of input features by using an xgboost algorithm, fits all the input features and EPO week dose by using the xgboost, sorts the average reduction of loss functions of each feature when the features are used as division attributes from large to small to obtain feature importance degree ranks, and uses the input features ranked at the front as the input features of the EPO week dose algorithm component;the EPO weekly dose algorithm component is composed of a decision regression tree, the decision regression tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold value of the input feature, the nodes are divided layer by layer according to the threshold values, the final output result is the mean value of numerical values in the nodes, and the mean value of the numerical values in the nodes is the EPO weekly dose;the dialysate flow characteristic importance degree analysis module performs characteristic importance degree analysis on input characteristics by using an xgboost algorithm, fits all the input characteristics and dialysate flow by using the xgboost, sorts the average reduction of loss functions of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance degree sequence, and takes the input characteristic with the top rank as the input characteristic of the dialysate flow algorithm module;the dialysate flow algorithm component is composed of a logistic regression algorithm, and firstly, input characteristics are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysate flow; the dialysate flow with the maximum probability is the output result of the dialysate flow algorithm component;the dialysate temperature feature importance analyzing component analyzes the feature importance of the input features by using an xgboost algorithm, fits all the input features and the dialysate temperature by using an xgboost, sorts the average reduction quantity of loss functions of each feature as a partition attribute from large to small to obtain a feature importance ranking, and takes the input feature with the top ranking as the input feature of the dialysate temperature algorithm component;the dialysate temperature algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely, the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate temperature parameter;the dialysate conductivity feature importance analysis component analyzes feature importance of input features by using an xgboost algorithm, all the input features and the dialysate conductivity are fitted by using the xgboost, the average reduction of loss functions of each feature when the feature is used as a partition attribute is sorted from large to small to obtain feature importance sorting, and the input feature with the top ranking is used as the input feature of the dialysate conductivity algorithm component;the dialysate conductivity algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely, the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the conductivity parameter of the dialysate;the dialysate potassium concentration characteristic importance degree analysis component performs characteristic importance degree analysis on input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate potassium concentration by using the xgboost, sorts the average reduction of loss functions of each characteristic when the characteristics are used as partition attributes from large to small to obtain a characteristic importance degree sort, and uses the input characteristic with the top rank as the input characteristic of the dialysate potassium concentration algorithm component;the dialysate potassium concentration algorithm component is composed of a classified random forest which is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate potassium concentration parameter;the dialysate sodium concentration characteristic importance degree analysis component performs characteristic importance degree analysis on input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate sodium concentration by using an xgboost, sorts the average reduction of loss functions of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance degree sort, and uses the input characteristic with the top rank as the input characteristic of the dialysate sodium concentration algorithm component;the dialysate sodium concentration algorithm component is composed of a BP neural network classification algorithm; the other structures of the BP neural network classification algorithm except the output layer are the same as the BP neural network regression algorithm, the output layer of the BP neural network classification algorithm is the classification category number of dialysate sodium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of the hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to the range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate sodium concentration category is obtained;the dialysate calcium concentration characteristic importance degree analysis component performs characteristic importance degree analysis on input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate calcium concentration by using an xgboost, sorts the average reduction of loss functions of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance degree sort, and uses the input characteristic with the top rank as the input characteristic of the dialysate calcium concentration algorithm component;the dialysate calcium concentration algorithm component is composed of a BP neural network classification algorithm; the other structures of the BP neural network classification algorithm except an output layer are the same as the BP neural network regression algorithm, the output layer of the BP neural network classification algorithm is the number of classification categories of dialysate calcium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of the hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to the range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate calcium concentration category is obtained.
- 4. A method of generating a hemodialysis protocol based on an AI hemodialysis protocol generation model, comprising: the AI hemodialysis scheme generation model comprises a dialyzer submodel, a dialysis setting submodel, a dialysis medicament submodel and a dialysate submodel; the method comprises the following steps:s1, inputting the primary input characteristics into a dialyzer submodel, wherein the dialyzer submodel gives dialysis mode parameter values and dialyzer model parameter values;s2, combining the primary input characteristics with dialysis mode parameter values and dialyzer model parameter values to generate secondary input characteristics;s3, inputting the secondary input characteristics into a dialysis setting submodel, and outputting a pre-dehydration value, dialysis time and dialysis blood flow volume by the dialysis setting submodel;s4, combining the primary input features with the pre-dehydration value, the dialysis time and the dialysis blood flow volume to generate three-time input features;s5, inputting the three input characteristics into a dialysis agent sub-model, wherein the dialysis agent sub-model gives anticoagulant dose and EPO weekly dose;s6, combining the primary input characteristics with the anticoagulant dose and the EPO weekly dose to generate four input characteristics;s7, inputting the four-time input characteristics into a dialysate sub-model, wherein the dialysate sub-model gives a dialysate flow, a dialysate temperature, a dialysate conductivity, a dialysate potassium concentration, a dialysate sodium concentration and a dialysate calcium concentration;s8, finally, combining a dialysis mode parameter value, a dialyzer model parameter value, a pre-dehydration value parameter, a dialysis time parameter, a dialysis blood flow parameter, an anticoagulant dosage parameter, an EPO weekly dosage parameter, a dialysate flow parameter, a dialysate temperature parameter, a dialysate conductivity parameter, a dialysate potassium concentration parameter, a dialysate sodium concentration parameter and a dialysate calcium concentration parameter by using the AI hemodialysis scheme generation model to generate a hemodialysis scheme;the primary input features are derived from a medical facility hemodialysis database, including basic information, pre-dialysis grid checks, historical dialysis protocols, and historical examination data.
- 5. The method of generating a hemodialysis protocol based on an AI hemodialysis protocol generation model according to claim 4, wherein:(1) The dialyzer submodel comprises a dialyzer model component and a dialysis mode component;the dialyzer model component comprises a dialyzer model algorithm component and a dialyzer model feature importance analysis component;the dialyzer model component gives the dialyzer model component input characteristics through the dialyzer model characteristic importance degree analysis component, inputs the dialyzer model component input characteristics into the dialyzer model algorithm component, and finally gives dialyzer model parameter values through the dialyzer model algorithm component;the dialysis mode component comprises a dialysis mode algorithm component and a dialysis mode feature importance analysis component;the dialysis mode component firstly gives the input characteristics of the dialysis mode component through the dialysis mode characteristic importance degree analysis component, and then inputs the input characteristics of the dialysis mode component into the dialysis mode algorithm component; finally, the dialysis mode algorithm component gives out dialysis mode parameter values;(2) A dialysis setting sub-model; the dialysis setting submodel comprises a preset dehydration component, a dialysis time component and a dialysis blood flow component;the preset dehydration component comprises a preset dehydration algorithm component and a preset dehydration characteristic importance degree analysis component;the preset dehydration component firstly gives the input characteristics of the preset dehydration component through the preset dehydration characteristic importance analysis component; then, inputting the input characteristics of the preset dehydration component into a preset dehydration algorithm component; finally, a preset dehydration algorithm component gives a preset dehydration value parameter;the dialysis time component comprises a dialysis time algorithm component and a dialysis time characteristic importance degree analysis component;the dialysis time component firstly gives the input characteristics of the dialysis time component through the dialysis time characteristic importance degree analysis component; then, inputting the input characteristics of the dialysis time component into a dialysis time algorithm component; finally, the dialysis time algorithm component gives out dialysis time parameters;the dialysis blood flow component comprises a dialysis blood flow algorithm component and a dialysis blood flow characteristic importance analysis component;firstly, the dialysis blood flow algorithm component gives the input characteristics of the dialysis blood flow component through a dialysis blood flow characteristic importance degree analysis component; then, inputting the input characteristics of the dialysis blood flow component into a dialysis blood flow algorithm component; finally, the dialysis blood flow algorithm component gives out dialysis blood flow parameters;(3) The dialysis agent sub-model comprises an anticoagulant component and an EPO weekly dose component;the anticoagulant component comprises an anticoagulant algorithm component and an anticoagulant characteristic importance analysis component;the anticoagulant component firstly gives the input characteristics of the anticoagulant component through the anticoagulant characteristic importance analysis component; then inputting the anticoagulant component input characteristics into an anticoagulant algorithm component; finally, the anticoagulant algorithm component gives anticoagulant dosage parameters;the EPO week dosage component comprises an EPO week dosage algorithm component and an EPO week dosage characteristic importance analysis component;the EPO week dosage component firstly gives EPO week dosage component input characteristics through an EPO week dosage characteristic importance analysis component; then, the EPO weekly dose component input features are input into the EPO weekly dose algorithm component; finally, the EPO weekly dose algorithm component gives EPO weekly dose parameters;(4) The dialysate submodel comprises a dialysate flow component, a dialysate temperature component, a dialysate conductivity component, a dialysate potassium concentration component, a dialysate sodium concentration component and a dialysate calcium concentration component;the dialysate flow component comprises a dialysate flow algorithm component and a dialysate flow characteristic importance analysis component;firstly, the dialysate flow component gives input characteristics of the dialysate flow component through a dialysate flow characteristic importance analysis component; then inputting the dialysate flow component input characteristics into a dialysate flow algorithm component; finally, the dialysate flow algorithm component gives dialysate flow parameters;the dialysate temperature component comprises a dialysate temperature algorithm component and a dialysate temperature feature importance analysis component;the dialysate temperature component firstly gives input characteristics of the dialysate temperature component through a dialysate temperature characteristic importance analysis component; then inputting the dialysate temperature component input characteristics into a dialysate temperature algorithm component; finally, the dialysate temperature algorithm component gives a dialysate temperature parameter;the dialysate conductivity component comprises a dialysate conductivity algorithm component and a dialysate conductivity feature importance analysis component;the dialysate conductivity component firstly gives input characteristics of the dialysate conductivity component through the dialysate conductivity characteristic importance degree analysis component; then, inputting the input characteristics of the dialysate conductivity component into a dialysate conductivity algorithm component; finally, the dialysate conductivity algorithm component gives parameters of the conductivity of the dialysate;the dialysate potassium concentration component comprises a dialysate potassium concentration algorithm component and a dialysate potassium concentration characteristic importance analysis component;firstly, the input characteristics of the dialysate potassium concentration component are given by the dialysate potassium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate potassium concentration component into a dialysate potassium concentration algorithm component; finally, a dialysate potassium concentration algorithm component gives a dialysate potassium concentration parameter;the dialysate sodium concentration component comprises a dialysate sodium concentration algorithm component and a dialysate sodium concentration characteristic importance analysis component;firstly, the dialysate sodium concentration component gives input characteristics of the dialysate sodium concentration component through a dialysate sodium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate sodium concentration component into a dialysate sodium concentration algorithm component; finally, the dialysate sodium concentration algorithm component gives parameters of dialysate sodium concentration;the dialysate calcium concentration component comprises a dialysate calcium concentration algorithm component and a dialysate calcium concentration characteristic importance analysis component;firstly, the dialysate calcium concentration component gives the input characteristics of the dialysate calcium concentration component through a dialysate calcium concentration characteristic importance analysis component; then, inputting the input characteristics of the dialysate calcium concentration component into a dialysate calcium concentration algorithm component; finally, the dialysate calcium concentration algorithm component provides a dialysate calcium concentration parameter.
- 6. The method of generating a hemodialysis protocol based on the AI hemodialysis protocol generation model of claim 5, wherein:the dialyzer model feature importance analyzing component performs feature importance analysis on input features by using an xgboost algorithm, fits all the input features and the actual dialyzer model by using the xgboost, ranks the average reduction of the loss function when each input feature is used as a partition attribute from large to small to obtain a feature importance ranking, and takes the input feature ranked at the front as the dialyzer model algorithm component input feature;the dialyzer model algorithm component is composed of a decision classification tree, the decision classification tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold of the input feature, the input features are classified layer by layer according to the thresholds, and finally the input features are classified into a certain dialyzer category;the dialysis mode feature importance degree analysis module performs feature importance degree analysis on input features by using an xgboost algorithm, fits all the input features and the dialysis mode by using the xgboost, ranks the average reduction amount of the loss function when each input feature is used as a partition attribute from large to small to obtain feature importance degree ranks, and takes the input feature ranked at the top as the input feature of the dialysis mode algorithm module;the dialysis mode algorithm component is composed of a logistic regression algorithm, and firstly, input features are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysis mode; the dialysis mode with the maximum probability is the output result of the dialysis mode algorithm component;the preset dehydration characteristic importance degree analysis component analyzes the input characteristics by using an xgboost algorithm, fits all the input characteristics and the actual preset dehydration value by using the xgboost, sorts the average reduction quantity of the loss function when each characteristic is used as a partition attribute from large to small to obtain characteristic importance degree sorting, and takes the input characteristic with the top ranking as the input characteristic of the preset dehydration algorithm component;the preset dehydration algorithm component is composed of a BP neural network regression algorithm, the BP neural network regression algorithm is composed of a plurality of layers of hidden layers and an output layer, each layer of hidden layers and the output layer are composed of a plurality of neurons, each neuron in the first hidden layer combines input features in a multiple regression equation mode, an equation result is subjected to nonlinear transformation through an activation function, then the neuron of each hidden layer performs multiple regression on data obtained by the previous layer, the equation result is subjected to nonlinear transformation through the activation function, the number of the neurons of the output layer is 1, only the previous numerical value is subjected to multiple linear regression, and the final value output by the multiple linear regression is the preset dehydration value predicted by the model;the dialysis time characteristic importance degree analysis module performs characteristic importance degree analysis on input characteristics by using an xgboost algorithm, fits all the input characteristics and dialysis time values by using the xgboost, sorts the average reduction of loss functions of each characteristic when the characteristic is used as a partition attribute from large to small to obtain characteristic importance degree sorting, and uses the input characteristic with the top ranking as the input characteristic of the dialysis time algorithm module;the dialysis time algorithm component is composed of classified random forests, each random forest is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the random forest is the dialysis time parameter;the dialysis blood flow characteristic importance degree analysis component analyzes the characteristic importance degree of the input characteristics by an xgboost algorithm, fits all the input characteristics and the dialysis blood flow value by the xgboost, sorts the average reduction quantity of the loss function of each characteristic when the characteristic is used as a division attribute from large to small to obtain characteristic importance degree sorting, and uses the input characteristic with the top ranking as the input characteristic of the dialysis blood flow algorithm component;the dialysis blood flow algorithm component is composed of a multiple linear regression equation, and the input characteristics are substituted into the multiple linear regression equation to obtain a final numerical value, namely a value of the blood flow parameter;the anticoagulant feature importance degree analysis component analyzes feature importance degrees of input features by using an xgboost algorithm, all the input features and anticoagulant doses are fitted by using the xgboost, loss function average reduction quantity of each feature when the feature is used as a partition attribute is sorted from large to small to obtain feature importance degree sorting, and the input feature with the top ranking is used as an anticoagulant algorithm component input feature;the anticoagulant dose algorithm component is composed of a decision regression tree algorithm, the decision regression tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold value of the input feature, the nodes are divided layer by layer according to the threshold values, the final output result is the mean value of numerical values in the node, and the mean value of the numerical values in the node is the anticoagulant dose;the EPO week dose feature importance analyzing component analyzes the feature importance of input features by an xgboost algorithm, fits all the input features and EPO week doses by the xgboost, sorts the average reduction of loss functions of each feature as a division attribute from large to small to obtain a feature importance ranking, and takes the input feature ranked at the front as the input feature of the EPO week dose algorithm component;the EPO weekly dose algorithm component is composed of a decision regression tree, the decision regression tree is composed of a plurality of nodes, each node is provided with an input feature and a threshold value of the input feature, the nodes are divided layer by layer according to the threshold values, the final output result is the mean value of numerical values in the nodes, and the mean value of the numerical values in the nodes is the EPO weekly dose;the dialysate flow characteristic importance degree analysis module performs characteristic importance degree analysis on input characteristics by using an xgboost algorithm, fits all the input characteristics and dialysate flow by using the xgboost, sorts the average reduction of loss functions of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance degree sequence, and takes the input characteristic with the top rank as the input characteristic of the dialysate flow algorithm module;the dialysate flow algorithm component is composed of a logistic regression algorithm, and firstly, input characteristics are substituted into the logistic regression algorithm to obtain a multiple regression equation; then, substituting the result of the multiple regression equation into the softmax function to obtain the probability of each dialysate flow; the dialysate flow with the maximum probability is the output result of the dialysate flow algorithm component;the dialysate temperature feature importance analyzing component analyzes feature importance of input features by using an xgboost algorithm, all the input features and the dialysate temperature are fitted by using the xgboost, loss function average reduction quantity of each feature is ranked from large to small when the feature is used as a partition attribute to obtain feature importance ranking, and the input feature ranked at the front is used as the input feature of the dialysate temperature algorithm component;the dialysate temperature algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely, the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate temperature parameter;the dialysate conductivity feature importance analysis component analyzes feature importance of input features by using an xgboost algorithm, all the input features and the dialysate conductivity are fitted by using the xgboost, the average reduction of loss functions of each feature when the feature is used as a partition attribute is sorted from large to small to obtain feature importance sorting, and the input feature with the top ranking is used as the input feature of the dialysate conductivity algorithm component;the dialysate conductivity algorithm component is composed of classified random forests; the classified random forest is composed of a plurality of decision trees, the final result adopts a voting method, namely, the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the conductivity parameter of the dialysate;the dialysate potassium concentration characteristic importance degree analysis component performs characteristic importance degree analysis on input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate potassium concentration by using the xgboost, sorts the average reduction of loss functions of each characteristic when the characteristics are used as partition attributes from large to small to obtain a characteristic importance degree sort, and uses the input characteristic with the top rank as the input characteristic of the dialysate potassium concentration algorithm component;the dialysate potassium concentration algorithm component is composed of a classified random forest which is composed of a plurality of decision trees, the final result adopts a voting method, the voting method is that the mode of the result obtained by each decision tree is taken as the final result of the random forest, and the final result of the classified random forest is the dialysate potassium concentration parameter;the dialysate sodium concentration characteristic importance degree analysis component performs characteristic importance degree analysis on input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate sodium concentration by using an xgboost, sorts the average reduction of loss functions of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance degree sort, and uses the input characteristic with the top rank as the input characteristic of the dialysate sodium concentration algorithm component;the dialysate sodium concentration algorithm component is composed of a BP neural network classification algorithm; the other structures of the BP neural network classification algorithm except the output layer are the same as the BP neural network regression algorithm, the output layer of the BP neural network classification algorithm is the classification category number of dialysate sodium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of the hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to the range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate sodium concentration category is obtained;the dialysate calcium concentration characteristic importance degree analysis component performs characteristic importance degree analysis on input characteristics by using an xgboost algorithm, fits all the input characteristics and the dialysate calcium concentration by using an xgboost, sorts the average reduction of loss functions of each characteristic when the characteristic is used as a partition attribute from large to small to obtain a characteristic importance degree sort, and uses the input characteristic with the top rank as the input characteristic of the dialysate calcium concentration algorithm component;the dialysate calcium concentration algorithm component is composed of a BP neural network classification algorithm; the other structures of the BP neural network classification algorithm except an output layer are the same as the BP neural network regression algorithm, the output layer of the BP neural network classification algorithm is the number of classification categories of dialysate calcium concentration algorithm components, each neuron of the output layer calculates the data of the last layer of the hidden layer through a multivariate regression equation to obtain a calculation result, the calculation result is subjected to a sigmoid function, the result is compressed to the range of 0 to 1, the result is regarded as the probability value of the category corresponding to the neuron, and finally the value of which neuron is the largest is obtained, so that the value of the corresponding dialysate calcium concentration category is obtained.
- 7. The method of generating a hemodialysis protocol based on an AI hemodialysis protocol generation model according to claim 4, wherein:the basic information data comprises age, sex, height, weight, blood type, nationality, dialysis age, primary morbidity, infection history and important medical history;the examination of the pre-dialysis lattice comprises dry body weight, last machine-off body weight, pre-dialysis body weight, blood pressure, heart rate and body temperature;the historical dialysis scheme comprises 13 indexes including a dialysis mode, a dialyzer, preset dehydration, dialysis time, anticoagulant dose, blood flow, EPO weekly dose, dialysate K concentration, dialysate Ca concentration, dialysate Na concentration, dialysate flow, conductivity and dialysate temperature;the historical test examination data includes hemoglobin, mean hemoglobin concentration, mean hemoglobin content, platelet count, white blood cell count, red blood cell specific volume, mean red blood cell volume, creatinine, urea, uric acid, albumin, total protein, prealbumin, glucose, sodium, potassium, calcium, phosphorus, chloride, bicarbonate, plasma fibrinogen, fibrin degradation products, plasma D-D dimer, serum troponin I, serum troponin T, myoglobin, serum iron, serum total iron binding capacity, serum ferritin, parathyroid hormone, occult blood, urine protein.
- 8. The method of generating a hemodialysis protocol based on the AI hemodialysis protocol generation model of claim 4, wherein: step S1 further comprises a sub-step of giving a primary input feature, comprising:s11, collecting characteristic data;acquiring existing hemodialysis overall process data from a hemodialysis database of a medical institution, and selecting dialysis data of a maintenance dialysis patient, wherein the dialysis data comprises basic information data, pre-dialysis grid inspection data, historical dialysis scheme data and historical inspection data;the basic information data comprises age, sex, height, weight, blood type, nationality, dialysis age, primary morbidity, infection history and important medical history;the pre-dialysis check data comprises dry body weight, last off-machine body weight, pre-dialysis body weight, blood pressure, heart rate and body temperature;the historical dialysis scheme data comprises 13 indexes including a dialysis mode, a dialyzer, preset dehydration, dialysis time, anticoagulant dose, blood flow, EPO weekly dose, dialysate K concentration, dialysate Ca concentration, dialysate Na concentration, dialysate flow, conductivity and dialysate temperature;the historical test examination data includes hemoglobin, mean hemoglobin concentration, mean hemoglobin content, platelet count, white blood cell count, red blood cell specific volume, mean red blood cell volume, creatinine, urea, uric acid, albumin, total protein, prealbumin, glucose, sodium, potassium, calcium, phosphorus, chloride, bicarbonate, plasma fibrinogen, fibrin degradation products, plasma D-D dimer, serum troponin I, serum troponin T, myoglobin, serum iron, serum total iron binding capacity, serum ferritin, parathyroid hormone, occult blood, urine protein;s12, cleaning the characteristic data to obtain standardized characteristic data; the characteristic data comprises numerical characteristic data and non-numerical characteristic data;s121, performing unit conversion on each type of numerical characteristic data, and converting each type of numerical characteristic data into numerical characteristic data of a unified unit in a unified unit;s122, processing abnormal values in numerical characteristic data of unified units;the abnormal value processing method comprises the following steps:s1221, if the numerical data in the numerical characteristic data of the unified unit obeys normal distribution, the data in the numerical characteristic data and the data except that the average value of the data in the numerical characteristic data is more than or less than 3 times of the standard deviation are abnormal values; then, eliminating the data of the row where the abnormal value is located; the average value refers to the average value of the columns of the numerical characteristic data;s1222, if the numerical characteristics in the numerical characteristic data of the unified unit do not comply with normal distribution, adopting a box-line graph method; the numerical characteristic data of the unified unit is smaller than Q1-1.5QR, or the numerical characteristic data of the unified unit is larger than Q3+1.5QR and is an abnormal value; then, eliminating the data of the row where the abnormal value is located;further, Q1 is a lower quartile, Q3 is an upper quartile, and QR is the subtraction of the upper quartile and the lower quartile;obtaining numerical characteristic data of a unified unit with abnormal values removed;s123, synchronizing, and assigning the non-numerical characteristic data by referring to a non-numerical data conversion rule to obtain assigned non-numerical characteristic data;s124, finally, forming standardized feature data by the numerical feature data of the unified unit with the abnormal values removed and the assigned non-numerical feature data;the standardized feature data is a one-time input feature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211418479.1A CN115579108B (en) | 2022-11-14 | 2022-11-14 | AI hemodialysis scheme model modeling method and method for generating hemodialysis scheme |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211418479.1A CN115579108B (en) | 2022-11-14 | 2022-11-14 | AI hemodialysis scheme model modeling method and method for generating hemodialysis scheme |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115579108A true CN115579108A (en) | 2023-01-06 |
| CN115579108B CN115579108B (en) | 2023-04-07 |
Family
ID=84588704
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211418479.1A Active CN115579108B (en) | 2022-11-14 | 2022-11-14 | AI hemodialysis scheme model modeling method and method for generating hemodialysis scheme |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115579108B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109686446A (en) * | 2019-01-22 | 2019-04-26 | 江苏易透健康科技有限公司 | A kind of hemodialysis program analysis method and system based on track planning of dual robots study |
| CN208823585U (en) * | 2017-11-07 | 2019-05-07 | 北京英福美信息科技股份有限公司 | The online Kt/V monitoring device of haemodialysis |
| CN112509669A (en) * | 2021-02-01 | 2021-03-16 | 肾泰网健康科技(南京)有限公司 | AI technology-based renal disease hemodialysis scheme customization method and system |
| CN113744857A (en) * | 2021-09-30 | 2021-12-03 | 深圳市裕辰医疗科技有限公司 | Hemodialysis quality control auxiliary system |
-
2022
- 2022-11-14 CN CN202211418479.1A patent/CN115579108B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN208823585U (en) * | 2017-11-07 | 2019-05-07 | 北京英福美信息科技股份有限公司 | The online Kt/V monitoring device of haemodialysis |
| CN109686446A (en) * | 2019-01-22 | 2019-04-26 | 江苏易透健康科技有限公司 | A kind of hemodialysis program analysis method and system based on track planning of dual robots study |
| CN112509669A (en) * | 2021-02-01 | 2021-03-16 | 肾泰网健康科技(南京)有限公司 | AI technology-based renal disease hemodialysis scheme customization method and system |
| CN113744857A (en) * | 2021-09-30 | 2021-12-03 | 深圳市裕辰医疗科技有限公司 | Hemodialysis quality control auxiliary system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115579108B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP2618864B1 (en) | Obtaining control settings for a dialysis machine | |
| CN112509669A (en) | AI technology-based renal disease hemodialysis scheme customization method and system | |
| US20140129250A1 (en) | Device and method for monitoring the treatment of a patient | |
| CN105007959B (en) | The device and method that treatment parameter for treating patient is provided | |
| CN113509610A (en) | Distributed multi-mode information perception multi-channel auxiliary intelligent closed-loop drug delivery robot | |
| CN116453706A (en) | Hemodialysis scheme making method and system based on reinforcement learning | |
| CN115579108B (en) | AI hemodialysis scheme model modeling method and method for generating hemodialysis scheme | |
| CN116808339A (en) | Early hypotension monitoring, prevention and control system in hemodialysis | |
| US20230036006A1 (en) | Artificial intelligence assisted home therapy settings for dialysis | |
| Cornalba et al. | Building a normative decision support system for clinical and operational risk management in hemodialysis | |
| Patra et al. | Automated micro insulin dispenser system based on the model predictive control algorithm | |
| US20230010826A1 (en) | Control of a therapeutic delivery system | |
| Rahim et al. | Fuzzy expert system for fluid management in general anesthesia | |
| Shie et al. | An integrated model using the Taguchi method and artificial neural network to improve artificial kidney solidification parameters | |
| CN113577476A (en) | Medical data analysis system based on information collection of breathing machine | |
| CN112957553A (en) | Intelligent monitoring system and method for complications of hemodialysis patients | |
| Azar | Multi-adaptive neural-fuzzy system as a novel predictor of in-vivo blood side dialyser urea clearance | |
| Ardi et al. | Decision Tree for Predicting the Mortality in Hemodialysis Patient with Diabetes | |
| CN117809833B (en) | Remote intelligent peritoneal dialysis system | |
| CN113436733B (en) | Characteristic construction method of hemodialysis scheme generation model based on fusion experience | |
| RU2758338C1 (en) | Method for choosing personalized antianemic therapy for patients with chronic renal failure older than 15 years | |
| Barbieri et al. | Anemia management in end stage renal disease patients undergoing dialysis: a comprehensive approach through machine learning techniques and mathematical modeling | |
| Yuan et al. | Interpretable Machine Learning-Based Risk Scoring with Individual and Ensemble Model Selection for Clinical Decision Making | |
| Xiang et al. | Causal Convolution-Transformer for Time Series Forecasting of Dry Weight in Hemodialysis Patients | |
| CN117497196A (en) | ICU patient invasive ventilation prediction method, ICU patient invasive ventilation prediction device, ICU patient invasive ventilation prediction system and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: Modeling Method for AI Hemodialysis Scheme Model and Method for Generating Hemodialysis Scheme Granted publication date: 20230407 Pledgee: Nanjing Bank Co.,Ltd. Nanjing North Branch Pledgor: Shentai Health Technology (Nanjing) Co.,Ltd. Registration number: Y2024980005062 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right |