CN112071431B - Clinical path automatic generation method and system based on deep learning and knowledge graph - Google Patents
Clinical path automatic generation method and system based on deep learning and knowledge graph Download PDFInfo
- Publication number
- CN112071431B CN112071431B CN202010989064.4A CN202010989064A CN112071431B CN 112071431 B CN112071431 B CN 112071431B CN 202010989064 A CN202010989064 A CN 202010989064A CN 112071431 B CN112071431 B CN 112071431B
- Authority
- CN
- China
- Prior art keywords
- intersection
- deep learning
- clinical path
- clinical
- patient
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000013135 deep learning Methods 0.000 title claims abstract description 32
- 239000003814 drug Substances 0.000 claims abstract description 56
- 238000013136 deep learning model Methods 0.000 claims abstract description 54
- 238000009826 distribution Methods 0.000 claims abstract description 13
- 230000003203 everyday effect Effects 0.000 claims abstract description 8
- 239000013598 vector Substances 0.000 claims description 62
- 201000010099 disease Diseases 0.000 claims description 28
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 28
- 238000003745 diagnosis Methods 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 19
- 238000012795 verification Methods 0.000 claims description 19
- 238000004590 computer program Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 12
- 238000003860 storage Methods 0.000 claims description 10
- 230000037361 pathway Effects 0.000 claims description 8
- 230000002457 bidirectional effect Effects 0.000 claims description 6
- 125000004122 cyclic group Chemical group 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 5
- 238000003062 neural network model Methods 0.000 claims description 5
- 238000007477 logistic regression Methods 0.000 claims description 4
- 238000012417 linear regression Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 9
- 238000011272 standard treatment Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 7
- 229940079593 drug Drugs 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000007689 inspection Methods 0.000 description 4
- 230000007170 pathology Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 210000004072 lung Anatomy 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 208000011580 syndromic disease Diseases 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 239000002246 antineoplastic agent Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229940044683 chemotherapy drug Drugs 0.000 description 1
- 210000001072 colon Anatomy 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- QOFFJEBXNKRSPX-ZDUSSCGKSA-N pemetrexed Chemical compound C1=N[C]2NC(N)=NC(=O)C2=C1CCC1=CC=C(C(=O)N[C@@H](CCC(O)=O)C(O)=O)C=C1 QOFFJEBXNKRSPX-ZDUSSCGKSA-N 0.000 description 1
- 229960005079 pemetrexed Drugs 0.000 description 1
- 238000002271 resection Methods 0.000 description 1
- 238000001356 surgical procedure Methods 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
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
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
- G06N3/08—Learning methods
-
- 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
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
Abstract
The invention provides a clinical path automatic generation method and system based on deep learning and knowledge graph. The method comprises the steps of inputting characteristic information of a patient into a first deep learning model and a second deep learning model respectively, and correspondingly recommending a first clinical path and a second clinical path; taking intersection sets of all orders contained in the first clinical path and the second clinical path, dividing the orders in the intersection sets according to categories, and forming a medical operation order recommendation set and a medicine order recommendation set; acquiring an intersection of the medical operation doctor's advice recommendation set and the medical operation alias set to obtain a first intersection; acquiring an intersection of the medicine doctor advice recommendation set and the medicine commodity name set to obtain a second intersection; and obtaining the execution time of each medical order in the first intersection and the second intersection through the second clinical path, and obtaining the medical order distribution of the first intersection and the second intersection corresponding to the clinical path every day/every stage.
Description
Technical Field
The invention belongs to the field of clinical data processing, and particularly relates to a method and a system for automatically generating a clinical path based on deep learning and knowledge-graph.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
At present, two methods for generating clinical paths exist, one is to directly use the clinical paths of various diseases with fixed standards, and the other is to simply analyze the existing diagnosis and treatment schemes of doctors to make the clinical paths of various diseases by using the traditional machine learning method.
The inventor finds that the clinical path uniformly formulated according to the fixed standard is not combined with the actual conditions of various hospitals, departments, doctors and patients, and cannot meet various conditions of the patients, so that the clinical path cannot meet the actual treatment requirement, and therefore, the clinical path is difficult to play a role in actual clinical work. Clinical paths formulated based on the traditional machine learning method are not accurate enough due to the fact that the analysis range is not wide enough and the model complexity is not enough, and the requirements of clinical use of doctors cannot be met.
Disclosure of Invention
In order to solve the problems, the invention provides a method and a system for automatically generating a clinical path based on deep learning and a knowledge graph, which are used for intelligently recommending a clinical path suitable for patients of the type based on the deep learning and the knowledge graph and standardized treatment data, so that the accuracy of the formulated clinical path can be improved.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
the first aspect of the invention provides a clinical path automatic generation method based on deep learning and knowledge-graph.
In one or more embodiments, a method for automatically generating a clinical path based on deep learning and knowledge-graph includes:
inputting the characteristic information of the patient into a first deep learning model and a second deep learning model respectively, and correspondingly recommending a first clinical path and a second clinical path; wherein the first clinical path contains only orders and the second clinical path contains orders and their execution times;
taking intersection sets of all orders contained in the first clinical path and the second clinical path, dividing the orders in the intersection sets according to categories, and forming a medical operation order recommendation set and a medicine order recommendation set;
acquiring an intersection of the medical operation doctor's advice recommendation set and the medical operation alias set to obtain a first intersection; acquiring an intersection of the medicine doctor advice recommendation set and the medicine commodity name set to obtain a second intersection; the medical operation alias set and the medicine commodity name set are found based on a standard clinical path corresponding to the disease diagnosis name of the patient in the knowledge graph and the patient characteristic information;
and obtaining the execution time of each medical order in the first intersection and the second intersection through the second clinical path, and obtaining the medical order distribution of the first intersection and the second intersection corresponding to the clinical path every day/every stage.
A second aspect of the present invention provides an automated clinical pathway generation system based on deep learning and knowledge-graph.
In one or more embodiments, a clinical path automatic generation system based on deep learning and knowledge-graph, comprising:
the clinical path recommending module is used for inputting the characteristic information of the patient into the first deep learning model and the second deep learning model respectively and correspondingly recommending a first clinical path and a second clinical path; wherein the first clinical path contains only orders and the second clinical path contains orders and their execution times;
the first taking and delivering module is used for taking all orders contained in the first clinical path and the second clinical path into an intersection set, dividing the orders in the intersection set according to categories and forming a medical operation order recommended set and a medicine order recommended set;
the second intersection taking module is used for taking the intersection of the medical operation medical advice recommendation set and the medical operation alias set to obtain a first intersection; acquiring an intersection of the medicine doctor advice recommendation set and the medicine commodity name set to obtain a second intersection; the medical operation alias set and the medicine commodity name set are found based on a standard clinical path corresponding to the disease diagnosis name of the patient in the knowledge graph and the patient characteristic information;
and the medical advice distribution module is used for obtaining the execution time of each medical advice in the first intersection and the second intersection through the second clinical path, and obtaining the medical advice distribution of the first intersection and the second intersection corresponding to the clinical path every day/every stage.
A third aspect of the present invention provides a computer-readable storage medium.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps in a method for automatically generating clinical paths based on deep learning and knowledge-graph as described above.
A fourth aspect of the invention provides a computer device.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in the deep learning and knowledge-graph based clinical path automatic generation method as described above when the program is executed.
Compared with the prior art, the invention has the beneficial effects that:
the invention inputs the characteristic information of the patient into a first deep learning model and a second deep learning model respectively, and correspondingly recommends a first clinical path and a second clinical path; the first clinical path only comprises medical advice, the second clinical path comprises medical advice and execution time thereof, patient characteristic information provided by doctors is comprehensively analyzed based on a deep learning technology, and clinical paths suitable for the patients are intelligently recommended based on standardized treatment data, so that the accuracy of the formulated clinical paths is greatly improved;
the invention also finds out based on the knowledge graph and the standard clinical path corresponding to the disease diagnosis name of the patient in the patient characteristic information, and the medical operation alias set and the medicine commodity name set respectively correspond to the medical operation medical advice recommendation set and the medicine commodity name set to obtain a first intersection and a second intersection; and acquiring the execution time of each doctor's advice in the first intersection and the second intersection through the second clinical path, obtaining the doctor's advice distribution of the first intersection and the second intersection corresponding to the clinical path every day/every stage, automatically recommending a common treatment scheme for a doctor for patients with certain diseases, providing the doctor with reference to the common treatment scheme when making the clinical path, making the whole treatment process for the clinical path, and fitting the practical situation of the hospital based on the history standard treatment data of the hospital, providing a feasible clinical path for the doctor, and greatly improving the practicability of the clinical path.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
FIG. 1 is a flow chart of a method for automatically generating clinical paths based on deep learning and knowledge-graph according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a first deep learning model of an embodiment of the present invention;
FIG. 3 is a schematic diagram of a second deep learning model of an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a clinical path automatic generation system based on deep learning and knowledge-graph according to an embodiment of the present invention.
Detailed Description
The invention will be further described with reference to the drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Term interpretation:
structural characteristics: for a certain feature of a patient, if the expression of the feature only needs structured data, the feature is called structured feature, and the doctor's advice information is structured feature information;
unstructured features: for a feature of a patient, if unstructured text data is required to express the feature, the feature is called unstructured feature, and the descriptive information of a disease symptom in the current medical history is unstructured feature information.
Medical operation standard name and medical operation alias: the medical operation class orders have both standard names and alias names, for example, "left upper lung lobe resection" is a standard name of a certain operation, and "resect left upper lung lobe" is an alias name of the operation, and one order has only one standard name and can have a plurality of alias names.
Pharmaceutical ingredient names and pharmaceutical commodity names: for drugs of different components, the component names thereof are recorded uniformly in a standard clinical path, but in clinical work of doctors, the name of a prescribed drug is often a commodity name of the drug instead of the component name, for example, "octenib" is a drug component name and "tylisha" is a drug commodity name of the component.
Knowledge graph S: unstructured feature information- > keywords and words, related data of the unstructured feature information refer to text parts describing the feature, the text parts contain keywords and words related to the index, and the knowledge graph S records keywords and words of different unstructured feature information.
Knowledge graph T: disease diagnosis name- > standard clinical path, all orders per day or per stage in its uniformly formulated standard clinical path are recorded for each disease.
Knowledge graph Y: the medicine component names- > the medicine commodity names, and for different medicine component names, different medicine commodity names corresponding to the medicine component names are recorded, wherein each medicine component name can contain a plurality of medicine commodity names.
Knowledge graph C: medical operation standard name- > medical operation alias, and recording different alias names corresponding to standard names of different medical operation orders.
Example 1
Referring to fig. 1, the method for automatically generating a clinical path based on deep learning and knowledge-graph of the present embodiment includes:
step S101: inputting the characteristic information of the patient into a first deep learning model and a second deep learning model respectively, and correspondingly recommending a first clinical path and a second clinical path; wherein the first clinical path contains only orders and the second clinical path contains orders and their execution times.
In an implementation, patient characteristic information provided by a physician is received, and this information is denoted as set A, and the disease diagnosis name in the patient characteristic is denoted as D.
The feature set A is respectively input into a first deep learning model M and a second deep learning model G, and the two models respectively output recommended clinical paths R_M and R_G. Wherein, the two clinical paths R_M and R_G are the full-flow treatment scheme, wherein R_G comprises the execution time of each order.
In a specific implementation, the first deep learning model M is used for receiving various characteristic information of a patient and pushing out a clinical path, namely a full-flow treatment scheme, suitable for the patient.
As shown in fig. 2, the first deep learning model M includes:
the bidirectional long-short-term memory cyclic neural network model is used for receiving unstructured text information word vectors of characteristic information of a patient, and outputting the result through a maximum pool to obtain a description vector representing the condition of the patient;
a logistic regression model for combining the vector of patient characteristic information structured information and the patient condition description vector, and mapping the combined vector to an index vector representing a similarity relationship between the patient information and all orders; wherein any element in the index vector represents a probability of the corresponding order entering the clinical path.
Specifically, unstructured text information word vectors (possibly multiple) of a patient p are input into a bidirectional long-short-time memory cyclic neural network model, and output results of all output word vectors are subjected to a maximum pool technology to obtain a vector C\u capable of representing patient p disease description p The method comprises the steps of carrying out a first treatment on the surface of the Vector L/u of structured information of patient p p And C/u p The combination is defined as vector V/u p ＝[L_ p ,C_ p ]。
Vector is obtained by using a logistic regression methodV_ p Mapping to an index vector ρ representing the similarity between the patient p's information and all orders, i.eWherein c1, c2 are parameter tensors in the logistic regression relationship, training is required to determine specific values, the dimension of ρ is the number of all possible orders, and the ith element ρ (i) in the vector ρ represents the probability of selecting the ith order to enter the clinical path. The cross entropy of ρ and the 0-1 probability distribution q of all order results is taken as the loss function of model M.
Before training of the first deep learning model M, the information vectors corresponding to 60%, 20% and 20% of the k patients corresponding to each disease in the standard treatment information base H are respectively divided into a training set, a verification set and a test set. The training set, the verification set and the test set are utilized to train the first deep learning model M.
The whole training process of the first deep learning model M is: storing the obtained model once every training preset times (for example, 100 times) on the training set; after the primary model is stored, the model is used for primary verification on a verification set, and the sum of the loss functions of all patients in the verification set is obtained and stored; after verifying for a preset number of times (for example, 100 times), selecting a parameter corresponding to a model with the minimum verification loss function sum as a final parameter of the first deep learning model M.
The construction process of the first deep learning model M is as follows:
step (a 1): the standard treatment information base H of the hospital is constructed, and the standard treatment information base H comprises years of history diagnosis and treatment data of the hospital, wherein the data comprise basic information, medical records first page, medical advice, inspection report, pathology report, electrocardio report and other information of different patients.
Specifically, the step (a 1) is specifically implemented as follows:
for each disease diagnosis name in the knowledge graph T, k (for example, k=100) "different patients" diagnosed as the disease are found in the hospital history diagnosis and treatment information base, and information such as basic information, medical history, medical records top page, medical orders, examination report, inspection report, pathology report, electrocardiographic report, and the like of these patients is stored in the information base H. The information base H has a storage structure as follows, and includes all information including disease diagnosis name- > patient number (k patients are counted for each syndrome name) - > basic information of a patient with a certain number, medical record top page, medical order, examination report, inspection report, pathology report, electrocardiograph report, and the like.
Wherein: the definition of "different patients" is that the similarity between the admission record text of each patient calculated using the Levenshtein distance is less than a certain threshold, such as 0.5.
For patient p, the information mainly includes two types, one is structured information (including sex, age, syndrome, disease diagnosis name of patient) and the other is unstructured text information (including medical record of patient, examination report, inspection report, etc.).
For the structured information, the structured information is converted into vector information (called a category vector) according to a 0-1 vector corresponding to the category, that is, the structured information can be represented by a certain vector, for example, as follows:
example 1: the sex of the patient is divided into two categories (male and female), the male category vector is (1, 0), the female category vector is (0, 1), and the sex information of the patient is converted from "male or female" to the vector "(1, 0) or (0, 1)".
Example 2: assuming that there are 3000 disease diagnosis names in the information base H, the class vector corresponding to the disease diagnosis name of the patient is 3000 dimensions, the dimension value corresponding to the disease diagnosis of the patient is 1, and the remaining dimension values are 0.
For patient p, all the structured information is converted into a category vector, and all the category vectors of the patient are combined into a vector, which is denoted as L\u p 。
It should be noted that, in other embodiments, the first deep learning model M may also be implemented in other existing structural forms, which will not be described here.
For unstructured text information, because the text information contains a large amount of irrelevant information, important relevant information in the text information is firstly screened and extracted (the information extraction method is shown in the following step iii), and then vectorization representation of the extracted text information is completed, namely, words in illness state description are represented by word vectors by using word segmentation technology (e.g. bargain segmentation) and word vector technology (e.g. word2vec, BERT and other methods).
The unstructured text information of a certain feature f is divided according to a colon, a comma, a semicolon and a period, and each small segment is called a small sentence. For each sentence, it is input into the third deep learning model k_f. For each unstructured feature f there is a corresponding third deep learning model k_f, determining if the sentence is extracted. Extracting all small sentences related to the feature f, splicing the small sentences into a text, and outputting the text.
The third deep learning model k_f has the following structure:
inputting a word vector set of a sentence into a Bi-gating cyclic network (Bi-GRU), obtaining a vector m_f from output results of all output word vectors through a maximum pool technique, and then performing linear regression mapping on the vector m_f to represent the probability p of extracting the sentence or not, namely, p=c1×m_f+c2, wherein c1 and c2 are parameter tensors in the linear regression relation, and training is needed to determine specific values. The cross entropy of p and the true 0-1 probability q of whether to extract the clause is taken as a loss function of the model K_f.
The application process of the third deep learning model K_f is as follows:
the standard treatment information base H of the hospital is obtained, a sentence is input into the deep learning model K_f, the probability p of the sentence being extracted is output, if the probability p is greater than a preset threshold (for example, 0.5 or other values), the sentence is judged to be extracted, and otherwise, the sentence is judged not to be extracted.
Specifically, the process of training the third deep learning model k_f is:
dividing word vector sets corresponding to 60%, 20% and 20% of all phrases of k patients corresponding to each disease in a standard treatment information base H into a training set, a verification set and a test set respectively;
the whole training process is as follows: storing the obtained model once every training preset times (for example, 100 times) on the training set; after the primary model is stored, the model is used for primary verification on a verification set, and the sum of the loss functions of all patients in the verification set is obtained and stored; after verifying for a preset number of times (for example, 100 times), selecting a parameter corresponding to a model with the minimum verification loss function sum as a final parameter of the third deep learning model K_f.
It should be noted that, in other embodiments, the third deep learning model k_f may also be implemented in other existing structural forms, which will not be described here.
Step (a 2): patient characteristic information given by a doctor is input into the deep learning model M, and probability vectors rho of all orders to be adopted are output.
Step (a 3): and selecting the doctor advice corresponding to the dimension with the probability larger than the preset probability value (such as 0.5) in the vector rho, otherwise, not selecting the doctor advice. Outputting all selected orders.
As shown in fig. 3, the second deep learning model G is used to receive various characteristic information of the patient, and to derive a clinical path suitable for the patient and an execution time of each order, that is, a full-flow treatment plan.
The second deep learning model G is an optimization model based on the first deep learning model M, and the construction process of the second deep learning model G is as follows:
step (b 1): constructing a standard treatment information base H of the hospital;
step (b 2): for each patient p in the standard treatment information base H, inputting the information into a trained model M, and vector V/u of model M p And establishing a vector library V corresponding to the information library H.
Specifically, unstructured text information word vectors (possibly multiple) of a patient p are input into a bidirectional long-short-time memory cyclic neural network model, and output results of all output word vectors are subjected to a maximum pool technology to obtain a vector C\u capable of representing patient p disease description p The method comprises the steps of carrying out a first treatment on the surface of the Vector L/u of structured information of patient p p And C/u p The combination is defined as vector V/u p ＝[L_ p ,C_ p ]。
Step (b 3): for the characteristic information of the patient u input by the doctor, inputting the characteristic information into the trained model M to obtain a vector V\u u 。
The unstructured text information word vector (possibly a plurality of word vectors) of the patient u is input into a bidirectional long-short-time memory cyclic neural network model, and the output results of all the output word vectors are subjected to the maximum pool technology to obtain a vector C\u capable of representing the illness state description of the patient u u The method comprises the steps of carrying out a first treatment on the surface of the Vector L/u of structured information of patient u u And C/u u The combination is defined as vector V/u u ＝[L_ u ,C_ u ]。
Solving the following optimization problems:
and taking all orders of the patient p corresponding to the optimal solution V_p of the optimization problem in the information base H and the execution time of each order as the output of the model G.
Step S102: and taking an intersection of all orders contained in the first clinical path and the second clinical path, dividing the orders in the intersection according to categories, and forming a medical operation order recommendation set and a medicine order recommendation set.
Specifically, the intersection of all orders contained in r_ M, R _g is obtained as a set B, and orders in the set B are classified into two types, one type is a medical operation order (for example, surgery, note that the medical operation order name herein may be an alias but not necessarily a standard name), the other type is a medicine order (for example, a chemotherapeutic drug pemetrexed, note that the medicine order name herein is a commodity name of a medicine but not necessarily a medicine component name), and the sets composed of two types of orders of medical operation and medicine in B are respectively denoted as b_1 and b_2.
Step S103: acquiring an intersection of the medical operation doctor's advice recommendation set and the medical operation alias set to obtain a first intersection; acquiring an intersection of the medicine doctor advice recommendation set and the medicine commodity name set to obtain a second intersection; the medical operation alias set and the medicine commodity name set are found based on the knowledge graph and the standard clinical path corresponding to the disease diagnosis name of the patient in the patient characteristic information.
Specifically, in the knowledge graph T, searching a standard clinical path corresponding to the disease diagnosis name D of the patient in the step 1, and respectively marking a set formed by two medical orders of medical operation and medicine as E_1_o and E_2_o; wherein, the medical operation name contained in e_1_o refers to the medical operation standard name, and the medicine contained in e_2_o refers to the medicine component name instead of the medicine commodity name.
For the medical operation standard names contained in E_1_o, searching the corresponding medical operation aliases in the knowledge graph C, and marking the set formed by all the found medical operation aliases as E_1.
For the medicine components contained in E_2_o, searching the corresponding medicine commodity names in the knowledge graph Y, and marking the set formed by all commodity names as E_2.
Intersection of B_1 and E_1 is taken, and a set C_1 is obtained; intersection of B_2 and E_2 is taken to obtain set C_2.
Step S104: and obtaining the execution time of each medical order in the first intersection and the second intersection through the second clinical path, and obtaining the medical order distribution of the first intersection and the second intersection corresponding to the clinical path every day/every stage.
Specifically, for each order in C_1, C_2, its execution time is obtained by R_G, from which an order allocation per day/stage of the clinical path C_1, C_2 is obtained.
According to the embodiment, the patient characteristic information provided by doctors is comprehensively analyzed based on the deep learning technology, and the clinical path suitable for the patients is intelligently recommended based on standardized treatment data, so that the accuracy of the formulated clinical path is greatly improved; searching and finding a medical operation alias set and a medicine commodity name set based on a standard clinical path corresponding to the disease diagnosis name of the patient in the knowledge graph and the patient characteristic information, and respectively correspondingly acquiring an intersection with the medical operation medical advice recommendation set and the medicine commodity name set to obtain a first intersection and a second intersection; and acquiring the execution time of each doctor's advice in the first intersection and the second intersection through the second clinical path, obtaining the doctor's advice distribution of the first intersection and the second intersection corresponding to the clinical path every day/every stage, automatically recommending a common treatment scheme for a doctor for patients with certain diseases, providing the doctor with reference to the common treatment scheme when making the clinical path, making the whole treatment process for the clinical path, and fitting the practical situation of the hospital based on the history standard treatment data of the hospital, providing a feasible clinical path for the doctor, and greatly improving the practicability of the clinical path.
Example two
Referring to fig. 4, the present embodiment provides a clinical path automatic generation system based on deep learning and knowledge-graph, which includes:
(1) The clinical path recommending module is used for inputting the characteristic clinical path recommending module of the patient, which is used for inputting the characteristic information of the patient into the first deep learning model and the second deep learning model respectively and correspondingly recommending a first clinical path and a second clinical path; wherein the first clinical path contains only orders and the second clinical path contains orders and their execution times;
(2) The first taking and delivering module is used for taking all orders contained in the first clinical path and the second clinical path into an intersection set, dividing the orders in the intersection set according to categories and forming a medical operation order recommended set and a medicine order recommended set;
(3) The second intersection taking module is used for taking the intersection of the medical operation medical advice recommendation set and the medical operation alias set to obtain a first intersection; acquiring an intersection of the medicine doctor advice recommendation set and the medicine commodity name set to obtain a second intersection; the medical operation alias set and the medicine commodity name set are found based on a standard clinical path corresponding to the disease diagnosis name of the patient in the knowledge graph and the patient characteristic information;
(4) And the medical advice distribution module is used for obtaining the execution time of each medical advice in the first intersection and the second intersection through the second clinical path, and obtaining the medical advice distribution of the first intersection and the second intersection corresponding to the clinical path every day/every stage.
The clinical path automatic generation system based on the deep learning and the knowledge graph of the present embodiment corresponds to the steps in the clinical path automatic generation method based on the deep learning and the knowledge graph of the first embodiment one by one, and the specific implementation process is as described in the first embodiment, which is not described here again.
Example III
The present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps in the deep learning and knowledge-graph-based clinical path automatic generation method described in the above embodiment.
Example IV
The present embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps in the method for automatically generating a clinical path based on deep learning and knowledge-graph according to the above embodiment when the processor executes the program.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random access Memory (Random AccessMemory, RAM), or the like.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. The automatic generation method of the clinical path based on the deep learning and the knowledge graph is characterized by comprising the following steps:
inputting the characteristic information of the patient into a first deep learning model and a second deep learning model respectively, and correspondingly recommending a first clinical path and a second clinical path; wherein the first clinical path contains only orders and the second clinical path contains orders and their execution times;
taking intersection sets of all orders contained in the first clinical path and the second clinical path, dividing the orders in the intersection sets according to categories, and forming a medical operation order recommendation set and a medicine order recommendation set;
acquiring an intersection of the medical operation doctor's advice recommendation set and the medical operation alias set to obtain a first intersection; acquiring an intersection of the medicine doctor advice recommendation set and the medicine commodity name set to obtain a second intersection; the medical operation alias set and the medicine commodity name set are found based on a standard clinical path corresponding to the disease diagnosis name of the patient in the knowledge graph and the patient characteristic information;
obtaining execution time of each medical order in the first intersection and the second intersection through the second clinical path, and obtaining medical order distribution of the first intersection and the second intersection corresponding to each day/each stage of the clinical path;
the whole training process of the first deep learning model is as follows: storing the obtained model once every training preset times on the training set; after the primary model is stored, the model is used for primary verification on a verification set, and the sum of the loss functions of all patients in the verification set is obtained and stored; after verifying the preset times, selecting the parameters corresponding to the model with the minimum total verification loss function as the final parameters of the first deep learning model M;
the second deep learning model is an optimization model based on the first deep learning model.
2. The automatic generation method of clinical pathway based on deep learning and knowledge graph according to claim 1, wherein the medical operation names contained in the standard clinical pathway are medical operation standard names, and the medicine names contained in the standard clinical pathway are medicine component names.
3. The method for automatically generating a clinical pathway based on deep learning and knowledge-graph of claim 1, wherein the patient characteristic information comprises structured information and unstructured text information.
4. The method for automatically generating a clinical path based on deep learning and knowledge-graph according to claim 1, wherein the first deep learning model comprises:
the bidirectional long-short-term memory cyclic neural network model is used for receiving unstructured text information word vectors of characteristic information of a patient, and outputting the result through a maximum pool to obtain a description vector representing the condition of the patient;
a logistic regression model for combining the vector of patient characteristic information structured information and the patient condition description vector, and mapping the combined vector to an index vector representing a similarity relationship between the patient information and all orders; wherein any element in the index vector represents a probability of the corresponding order entering the clinical path.
5. The method for automatically generating a clinical pathway based on deep learning and knowledge-graph as claimed in claim 4, wherein unstructured text information word vectors of patient characteristic information are extracted by a third deep learning model.
6. The method for automatically generating a clinical pathway based on deep learning and knowledge-graph according to claim 5, wherein the third deep learning model comprises:
the bidirectional gating loop network is used for receiving a word vector set of a sentence, and enabling output results of all output word vectors to pass through a maximum pool to obtain a maximum pooled vector;
a linear regression model for linearly mapping the maximum pooling vector onto a probability representing whether or not to extract the phrase.
7. The method for automatically generating clinical paths based on deep learning and knowledge-graph according to claim 1, wherein the second deep learning model is an optimization model based on the first deep learning model.
8. A clinical pathway automatic generation system based on deep learning and knowledge-graph, comprising:
the clinical path recommending module is used for inputting the characteristic information of the patient into the first deep learning model and the second deep learning model respectively and correspondingly recommending a first clinical path and a second clinical path; wherein the first clinical path contains only orders and the second clinical path contains orders and their execution times;
the first taking and delivering module is used for taking all orders contained in the first clinical path and the second clinical path into an intersection set, dividing the orders in the intersection set according to categories and forming a medical operation order recommended set and a medicine order recommended set;
the second intersection taking module is used for taking the intersection of the medical operation medical advice recommendation set and the medical operation alias set to obtain a first intersection; acquiring an intersection of the medicine doctor advice recommendation set and the medicine commodity name set to obtain a second intersection; the medical operation alias set and the medicine commodity name set are found based on a standard clinical path corresponding to the disease diagnosis name of the patient in the knowledge graph and the patient characteristic information;
the medical advice distribution module is used for obtaining the execution time of each medical advice in the first intersection and the second intersection through the second clinical path, and obtaining the medical advice distribution of the first intersection and the second intersection corresponding to the clinical path every day/every stage;
the whole training process of the first deep learning model is as follows: storing the obtained model once every training preset times on the training set; after the primary model is stored, the model is used for primary verification on a verification set, and the sum of the loss functions of all patients in the verification set is obtained and stored; after verifying the preset times, selecting the parameters corresponding to the model with the minimum total verification loss function as the final parameters of the first deep learning model M;
the second deep learning model is an optimization model based on the first deep learning model.
9. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps in the method for automatically generating clinical paths based on deep learning and knowledge-graph according to any one of claims 1-7.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the program, implements the steps in the deep learning and knowledge-graph based clinical path automatic generation method of any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010989064.4A CN112071431B (en) | 2020-09-18 | 2020-09-18 | Clinical path automatic generation method and system based on deep learning and knowledge graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010989064.4A CN112071431B (en) | 2020-09-18 | 2020-09-18 | Clinical path automatic generation method and system based on deep learning and knowledge graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112071431A CN112071431A (en) | 2020-12-11 |
| CN112071431B true CN112071431B (en) | 2024-02-23 |
Family
ID=73681228
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010989064.4A Active CN112071431B (en) | 2020-09-18 | 2020-09-18 | Clinical path automatic generation method and system based on deep learning and knowledge graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112071431B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116994704B (en) * | 2023-09-22 | 2023-12-15 | 北斗云方(北京)健康科技有限公司 | Reasonable medication discrimination method based on clinical multi-modal data deep representation learning |
| CN117542535B (en) * | 2024-01-10 | 2024-03-22 | 电子科技大学 | Clinical path correction method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109378064A (en) * | 2018-10-29 | 2019-02-22 | 南京医基云医疗数据研究院有限公司 | Medical data processing method, device electronic equipment and computer-readable medium |
| CN109727680A (en) * | 2018-12-28 | 2019-05-07 | 上海列顿信息科技有限公司 | A kind of region clinical path management system based on big data technology |
| KR20200022110A (en) * | 2018-08-22 | 2020-03-03 | 주식회사 위담바이오 | Oriental medicine Clinical data collection and deep learning based data analysis system |
| CN111191020A (en) * | 2019-12-27 | 2020-05-22 | 江苏省人民医院（南京医科大学第一附属医院） | Prescription recommendation method and system based on machine learning and knowledge graph |
-
2020
- 2020-09-18 CN CN202010989064.4A patent/CN112071431B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20200022110A (en) * | 2018-08-22 | 2020-03-03 | 주식회사 위담바이오 | Oriental medicine Clinical data collection and deep learning based data analysis system |
| CN109378064A (en) * | 2018-10-29 | 2019-02-22 | 南京医基云医疗数据研究院有限公司 | Medical data processing method, device electronic equipment and computer-readable medium |
| CN109727680A (en) * | 2018-12-28 | 2019-05-07 | 上海列顿信息科技有限公司 | A kind of region clinical path management system based on big data technology |
| CN111191020A (en) * | 2019-12-27 | 2020-05-22 | 江苏省人民医院（南京医科大学第一附属医院） | Prescription recommendation method and system based on machine learning and knowledge graph |
Non-Patent Citations (1)
| Title |
|---|
| 基于EMR的乳腺肿瘤知识图谱构建研究;崔洁;陈德华;乐嘉锦;;计算机应用与软件(12);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112071431A (en) | 2020-12-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111986770B (en) | Prescription medication auditing method, device, equipment and storage medium | |
| CN109920508B (en) | Prescription auditing method and system | |
| EP3234823B1 (en) | Differential medical diagnosis apparatus adapted in order to determine an optimal sequence of diagnostic tests for identifying a pathology by adopting diagnostic appropriateness criteria | |
| Ghosh et al. | Quro: facilitating user symptom check using a personalised chatbot-oriented dialogue system | |
| JP7304960B2 (en) | Health-informed prognostic score | |
| CN111048167B (en) | Hierarchical case structuring method and system | |
| US20140149132A1 (en) | Adaptive medical documentation and document management | |
| CN110827941A (en) | Electronic medical record information correction method and system | |
| CN112071431B (en) | Clinical path automatic generation method and system based on deep learning and knowledge graph | |
| CN113688255A (en) | Knowledge graph construction method based on Chinese electronic medical record | |
| CN114330267A (en) | Structural report template design method based on semantic association | |
| Bonin et al. | HBCP corpus: a new resource for the analysis of behavioural change intervention reports | |
| CN112259232B (en) | VTE risk automatic evaluation system based on deep learning | |
| JP2022504508A (en) | Systems and methods for model-assisted event prediction | |
| CN113539414A (en) | Method and system for predicting rationality of antibiotic medication | |
| Abaho et al. | Detect and Classify--Joint Span Detection and Classification for Health Outcomes | |
| Chen et al. | Automatically structuring on Chinese ultrasound report of cerebrovascular diseases via natural language processing | |
| CN112635072A (en) | ICU (intensive care unit) similar case retrieval method and system based on similarity calculation and storage medium | |
| CN110473636B (en) | Intelligent medical advice recommendation method and system based on deep learning | |
| EP3937105A1 (en) | Methods and systems for user data processing | |
| Memarzadeh et al. | A graph database approach for temporal modeling of disease progression | |
| CN113761899A (en) | Medical text generation method, device, equipment and storage medium | |
| CN113057588A (en) | Disease early warning method, device, equipment and medium | |
| Oliveira et al. | Towards an intelligent systems to predict nosocomial infections in intensive care | |
| Kim et al. | Automatic Diagnosis of medical conditions using Deep Learning with Symptom2Vec |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: No. 1237 Yingxiu Road, High tech Zone, Jinan City, Shandong Province, 250101 Applicant after: Zhongyang Health Technology Group Co.,Ltd. Address before: 12 / F, building 1, Aosheng building, 1166 Xinluo street, hi tech Zone, Jinan City, Shandong Province Applicant before: SHANDONG MSUNHEALTH TECHNOLOGY GROUP Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |