CN113171059B - Postoperative END risk early warning and related equipment based on multimodal monitoring information - Google Patents
Postoperative END risk early warning and related equipment based on multimodal monitoring information Download PDFInfo
- Publication number
- CN113171059B CN113171059B CN202110447993.7A CN202110447993A CN113171059B CN 113171059 B CN113171059 B CN 113171059B CN 202110447993 A CN202110447993 A CN 202110447993A CN 113171059 B CN113171059 B CN 113171059B
- Authority
- CN
- China
- Prior art keywords
- data
- time period
- clinical data
- early warning
- clinical
- 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
- 238000012544 monitoring process Methods 0.000 title claims abstract description 73
- 230000002980 postoperative effect Effects 0.000 title claims abstract description 42
- 238000000034 method Methods 0.000 claims abstract description 55
- 238000003070 Statistical process control Methods 0.000 claims abstract description 21
- 238000012545 processing Methods 0.000 claims description 30
- 238000012549 training Methods 0.000 claims description 22
- 238000012795 verification Methods 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 230000002537 thrombolytic effect Effects 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 8
- 238000001356 surgical procedure Methods 0.000 claims description 8
- 238000004458 analytical method Methods 0.000 claims description 7
- 238000004140 cleaning Methods 0.000 claims description 7
- 238000003384 imaging method Methods 0.000 claims description 7
- 206010061216 Infarction Diseases 0.000 claims description 6
- 239000003146 anticoagulant agent Substances 0.000 claims description 6
- 230000007574 infarction Effects 0.000 claims description 6
- 230000036772 blood pressure Effects 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 5
- 229940079593 drug Drugs 0.000 claims description 5
- 239000003814 drug Substances 0.000 claims description 5
- 229910003798 SPO2 Inorganic materials 0.000 claims description 4
- 101100478210 Schizosaccharomyces pombe (strain 972 / ATCC 24843) spo2 gene Proteins 0.000 claims description 4
- 210000004369 blood Anatomy 0.000 claims description 4
- 239000008280 blood Substances 0.000 claims description 4
- 238000001990 intravenous administration Methods 0.000 claims description 4
- 238000010845 search algorithm Methods 0.000 claims description 4
- 206010002091 Anaesthesia Diseases 0.000 claims description 3
- 206010003658 Atrial Fibrillation Diseases 0.000 claims description 3
- 206010069729 Collateral circulation Diseases 0.000 claims description 3
- 238000012276 Endovascular treatment Methods 0.000 claims description 3
- 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 3
- 206010020772 Hypertension Diseases 0.000 claims description 3
- 208000031481 Pathologic Constriction Diseases 0.000 claims description 3
- 230000037005 anaesthesia Effects 0.000 claims description 3
- 230000000702 anti-platelet effect Effects 0.000 claims description 3
- 230000010100 anticoagulation Effects 0.000 claims description 3
- 238000004820 blood count Methods 0.000 claims description 3
- 208000029078 coronary artery disease Diseases 0.000 claims description 3
- 239000008103 glucose Substances 0.000 claims description 3
- 210000000265 leukocyte Anatomy 0.000 claims description 3
- 210000000440 neutrophil Anatomy 0.000 claims description 3
- 230000036262 stenosis Effects 0.000 claims description 3
- 208000037804 stenosis Diseases 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 2
- 101000637031 Homo sapiens Trafficking protein particle complex subunit 9 Proteins 0.000 claims 1
- 108010089417 Sex Hormone-Binding Globulin Proteins 0.000 claims 1
- 102100031926 Trafficking protein particle complex subunit 9 Human genes 0.000 claims 1
- 206010072731 White matter lesion Diseases 0.000 claims 1
- 230000001537 neural effect Effects 0.000 claims 1
- 238000013151 thrombectomy Methods 0.000 claims 1
- 230000006870 function Effects 0.000 description 18
- 239000000523 sample Substances 0.000 description 17
- 238000011282 treatment Methods 0.000 description 12
- 238000013528 artificial neural network Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 6
- 230000006866 deterioration Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000029058 respiratory gaseous exchange Effects 0.000 description 5
- 230000002159 abnormal effect Effects 0.000 description 4
- 230000004913 activation Effects 0.000 description 4
- 230000004872 arterial blood pressure Effects 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 210000005036 nerve Anatomy 0.000 description 4
- 238000002560 therapeutic procedure Methods 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 3
- 238000012937 correction Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000010837 poor prognosis Methods 0.000 description 3
- 238000002604 ultrasonography Methods 0.000 description 3
- 208000028399 Critical Illness Diseases 0.000 description 2
- 208000032843 Hemorrhage Diseases 0.000 description 2
- 208000032382 Ischaemic stroke Diseases 0.000 description 2
- 208000034800 Leukoencephalopathies Diseases 0.000 description 2
- 208000006011 Stroke Diseases 0.000 description 2
- 229940127219 anticoagulant drug Drugs 0.000 description 2
- 238000013176 antiplatelet therapy Methods 0.000 description 2
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 210000004204 blood vessel Anatomy 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000003727 cerebral blood flow Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 239000006185 dispersion Substances 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000000926 neurological effect Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 229910052760 oxygen Inorganic materials 0.000 description 2
- 239000001301 oxygen Substances 0.000 description 2
- 230000000250 revascularization Effects 0.000 description 2
- 238000012552 review Methods 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000005406 washing Methods 0.000 description 2
- 206010008111 Cerebral haemorrhage Diseases 0.000 description 1
- 208000002403 Encephalocele Diseases 0.000 description 1
- 206010038563 Reocclusion Diseases 0.000 description 1
- 230000001154 acute effect Effects 0.000 description 1
- 210000000709 aorta Anatomy 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013481 data capture Methods 0.000 description 1
- 230000006837 decompression Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000001787 epileptiform Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000005713 exacerbation Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000036403 neuro physiology Effects 0.000 description 1
- 230000007658 neurological function Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000009885 systemic effect Effects 0.000 description 1
- 230000002087 whitening effect Effects 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7271—Specific aspects of physiological measurement analysis
- A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/40—Detecting, measuring or recording for evaluating the nervous system
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Heart & Thoracic Surgery (AREA)
- Public Health (AREA)
- Physiology (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Veterinary Medicine (AREA)
- Biomedical Technology (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Neurology (AREA)
- Neurosurgery (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
本发明涉及多模态监测信息的术后END风险预警及相关设备。设备包括：存储器和处理器；所述处理器用于执行以下操作：获取待测用户在当前时间周期的临床数据；将所述临床数据输入基于卷积理论的深度自编码器，得到所述临床数据对应的特征数据；利用统计过程控制方法获取所述临床数据的统计值，将所述临床数据的统计值与当前时间周期对应的预警阈值进行比较，根据比较结果确实现下一时间周期的术后END风险预警。本申请实现了实时更新术后END预警。
The invention relates to post-operative END risk early warning and related equipment for multimodal monitoring information. The device includes: a memory and a processor; the processor is configured to perform the following operations: acquiring clinical data of a user to be tested in a current time period; inputting the clinical data into a deep autoencoder based on convolution theory to obtain the clinical data Corresponding characteristic data; use the statistical process control method to obtain the statistical value of the clinical data, compare the statistical value of the clinical data with the early warning threshold corresponding to the current time period, and realize the postoperative END of the next time period according to the comparison result. Risk Warning. This application realizes real-time update of postoperative END early warning.
Description
Technical Field
The invention relates to the technical field of biological information, in particular to an automatic early warning, electronic equipment and a computer-readable storage medium for early nerve function deterioration after acute ischemic stroke intravascular treatment based on multi-modal monitoring information.
Background
Although current intravascular treatment has achieved the highest treatment recommendations of international organizations and international guidelines worldwide. Not all patients benefit from this, and about 40% of patients receiving endovascular treatment eventually develop a poor prognosis after surgery. Many patients experience an unintended exacerbation after surgery, eventually manifesting as a poor prognosis. Early Neurological Deterioration (END) is the most common and representative adverse event following endovascular therapy. END, which is currently most widely used, is defined as the deterioration of nerve function that occurs within 24 hours after revascularization, and the NIHSS score increases by 4 points or more from baseline. END symptomatic cerebral hemorrhage, reocclusion, ineffective recanalization, cerebral hernia, etc. are common causes of END occurrence, but the specific occurrence mechanism is still unclear. The occurrence of END is predicted in time, and positive corresponding intervention measures are taken in advance, so that the outcome of the patient can be reversed. Early neurological deterioration refers to improvement and deterioration of neurological function occurring in a short period after a patient receives revascularization treatment (including intravenous thrombolysis and intravascular treatment), and finally shows poor prognosis and even death.
Clinically, some END patients after endovascular treatment can not only have obvious abnormality in nerve function monitoring, but also have characteristic basic sign signal change, and are closely related to SND (sudden deterioration of nerve function). Changes in these complex physiological signals sometimes indicate the time of occurrence of END. It is noted that the sensitivity of these abnormal physiological monitoring signals, while significant, is not specific to the prediction of END, and many critically ill patients will similarly change due to other systemic disease problems. Based on the prior art, the current END prediction research result shows that the situation and clinical image information based on preoperative operation or electrophysiological monitoring information alone cannot be comprehensively predicted well. The real-time prediction of the END after the endovascular therapy can be better guided by simultaneously integrating the postoperative continuous multimode monitoring information based on the static clinical information and the image information. However, in actual clinical settings, these complex information, multimodal monitoring data, and other multi-modality data are monitored in real time and recorded only in limited amounts, but only a small portion of them are screened for clinical analysis. Meanwhile, the multidimensional and complex data can be partially analyzed by senior physicians through years of clinical experience, and misjudgment often occurs. In addition, the individuation trend of severe patients is obvious, the evaluation standard based on common patients is not applicable sometimes, and the mass data are difficult to judge by using the unified standard, so that the difficulty of analysis is further aggravated.
Disclosure of Invention
In view of the above problems, the present invention extracts complex multidimensional information and time-space representation of electrophysiological signals by using a deep self-encoder based on the convolution theory, and monitors the extracted electrophysiological characteristics by using a Statistical Process Control (SPC) method to achieve real-time update of END warning.
A postoperative END risk pre-warning device based on multimodal monitoring information, the device comprising: a memory and a processor;
the memory is to store program instructions;
the processor is configured to invoke program instructions that, when executed, are configured to:
acquiring clinical data of a user to be detected in a current time period, wherein the clinical data comprises baseline data, dynamic monitoring data obtained in the current time period and dynamic monitoring data obtained in a historical time period;
inputting the clinical data into a depth self-encoder based on a convolution theory to obtain characteristic data corresponding to the clinical data;
acquiring a statistical value of the clinical data by using a statistical process control method;
comparing the statistical value of the clinical data with an early warning threshold value corresponding to the current time period, and determining postoperative END risk early warning of the next time period according to the comparison result;
the training mode of the depth self-encoder based on the convolution theory and the determination mode of the early warning threshold value of each time period comprise the following steps:
acquiring clinical data of a patient without END after operation in each time period as a training sample, and acquiring clinical data of the patient with END after operation in each time period as a verification sample;
training a depth self-encoder based on a convolution theory by using a training sample; and verifying the depth self-encoder based on the convolution theory by using the verification sample, and processing the input data and the output data of the verification sample in each time period by using a statistical process control method to obtain the early warning threshold corresponding to each time period.
The baseline data comprises characteristic variables of baseline information, past history information, baseline laboratory indexes, baseline image information, intraoperative conditions and the like;
further, the baseline information (B) includes age, gender, baseline NIHSS score, baseline blood pressure, time to onset; the previous history information (H) comprises hypertension, coronary heart disease, atrial fibrillation and previous stroke, and the previous medication history (antiplatelet therapy and anticoagulant therapy); baseline laboratory indices (L) include admission blood glucose, white blood cell count, neutrophil count, PLT count, LDL); the baseline image information (I) comprises infarct volume, Mismatch volume, ASPECT score, infarct position, responsible blood vessel, occlusion degree, early signs, collateral circulation, leukoencephalopathy degree, microhemorrhage condition and the like; the intraoperative conditions (E) include intravenous thrombolysis treatment, anesthesia mode, intravascular treatment mode (stent removal, aspiration thrombolysis, stent formation, etc.), number of thrombolysis removal, intraoperative anticoagulation, antiplatelet application, postoperative TICI grading, residual stenosis, postoperative NIHSS score, time of surgery, etc.
Further, the dynamic monitoring data includes ECG (e), RESP (r), NIBP (nb), ABP (ab), SPO2(s), PULSE (p), EEG (eeg), TCD (tcd), etc.
Furthermore, data cleaning is carried out before feature extraction is carried out on clinical data, and the data cleaning comprises missing data processing and discrete and noise data processing.
The data cleansing includes processing missing data and processing discrete and noisy data. The processing of missing data can be done based on clinical analysis and model parameter optimization; the processing of the discrete and noise data can adopt a Binning method to process;
preferably, after the data washing is performed on the clinical data, all the data are normalized.
Preferably, a depth self-encoder based on convolution theory performs feature extraction on high-dimensional baseline data and high-data-volume dynamic monitoring data of a postoperative patient group without END and a postoperative patient group with END;
preferably, feature extraction is performed on high-dimensional baseline data and high-data-volume dynamic monitoring data through a depth self-encoder based on a convolution theory, and then integratable correction information encoding is obtained.
Preferably, the feature extraction is performed on the clinical data through a depth self-encoder based on the convolution theory, and then the integratable correction information encoding is obtained.
The Deep Auto-encoder (DAE) is mainly used for completing conversion learning task and can complete
Unsupervised learning and nonlinear feature extraction. The basic idea is to directly use one or more layers of neural networks to input
The data is mapped to obtain an output vector as a feature extracted from the input data (see figure). DAE is a utilization without supervision
Method for extracting high-dimensional complex input data from non-standard data through multi-layer nonlinear network for supervising and pre-training and systematic parameter optimization
And (5) layering the features, and obtaining a deep learning neural network structure represented by the distributed features of the original data. DAE is decoded by an encoder
A coder and a hidden layer.
Common DAEs cannot effectively solve the problems of pooling and whitening in complex data, and a large number of redundant parameters are forced to participate in calculation, so that the operation efficiency is low, while DAEs based on convolution theory can be used for processing multi-modal data in the application, the structure utilizes important local features to reconstruct original data, and all local features of input data share a weight matrix, so that a hidden layer of the DAE can completely store edge features limited by local space. The DAE based on the convolution theory is used for fitting the multi-modal data signal by using the linear combination of the basic modules, so that the speed and the accuracy of signal comprehensive identification are obviously improved.
Preferably, a heuristic search algorithm is used to optimize the autoencoder structure.
Preferably, a heuristic search algorithm is adopted to select the structure and the hyper-parameters of the automatic early warning model, and the automatic early warning model is subjected to optimization Model optimization。
Preferably, an exponential Weighted Moving-a verage (EWMA) in statistical process control is used for processing to obtain an early warning threshold corresponding to each time period, micro deviation of the multi-modal monitoring signal is detected, and automatic early warning is performed when index variation is monitored;
further, the larger the value output by the depth self-coder model, the greater the probability that END will occur.
Preferably, the feature extraction is carried out on the clinical data of a patient group with END after operation, the depth self-encoder model is improved, monitoring abnormal parameters are obtained, and the occurrence of END after operation is prompted;
preferably, the parameter RE is defined as a monitoring abnormality parameter, and the value of REs (i is 1 to N) is calculated by a trained depth encoder model, and the larger the value is, the higher the probability of occurrence of END is.
Preferably, an exponentially weighted moving average control map (EWMA) statistic of the statistical process control is calculated to obtain a weighted average of the sample means.
Preferably, according to the verification sample data, calculating the exponential weighted moving average control chart statistic of statistical process control to obtain the early warning threshold value.
Preferably, clinical data of the patients with END after operation and the patients without END after operation in each time period are obtained as verification samples to improve the early warning threshold. And further optimizing an early warning threshold value through normal and abnormal sample data.
Preferably, the early warning threshold corresponding to each time period is different.
EMWA determines the predicted value by giving different weights to the observed value, finding the moving average value according to the different weights
A method. The method adopts EMWA to better meet the change characteristics of human biological signals, and can combine the baseline condition and the baseline condition of a patient
The basic monitoring indexes can meet the requirement that the recent observation values of the observation and monitoring indexes have larger influence on the predicted values and can reflect the recent observation values
Trend of phase change. The intelligent understanding problem of the mutation signal in the multimode monitoring of the critically ill patients is solved by applying the technology.
A postoperative END risk early warning method based on multi-modal monitoring information comprises the following steps:
acquiring clinical data of a user to be detected in a current time period, wherein the clinical data comprises baseline data, dynamic monitoring data obtained in the current time period and dynamic monitoring data obtained in a historical time period;
inputting the clinical data into a depth self-encoder based on a convolution theory to obtain characteristic data corresponding to the clinical data;
acquiring a statistical value of the clinical data by using a statistical process control method;
and comparing the statistical value of the clinical data with an early warning threshold value corresponding to the current time period, and determining postoperative END risk early warning of the next time period according to a comparison result.
A postoperative END risk early warning device based on multi-modal monitoring information includes:
the system comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring clinical data of a user to be detected in a current time period, and the clinical data comprises baseline data, dynamic monitoring data obtained in the current time period and dynamic monitoring data obtained in a historical time period;
the processing unit is used for inputting the clinical data into a depth self-encoder based on a convolution theory to obtain characteristic data corresponding to the clinical data and acquiring a statistical value of the clinical data by using a statistical process control method;
and the prediction unit is used for comparing the statistic value of the clinical data with an early warning threshold value corresponding to the current time period and determining postoperative END risk early warning of the next time period according to a comparison result.
A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the post-operative END risk pre-warning method described above.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a schematic flow chart of a postoperative END risk early warning method based on multi-modal monitoring information according to an embodiment of the present invention;
FIG. 2 is a schematic block diagram of a post-operation END risk early warning device based on multi-modal monitoring information according to an embodiment of the present invention;
FIG. 3 is a diagram of an internal structure of a depth self-encoder according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of model training and improvement provided by an embodiment of the present invention;
fig. 5 is a schematic diagram of the construction and early warning model provided in the embodiment of the present invention;
fig. 6 is a schematic view of bedside multimodal monitoring.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention.
In some of the flows described in the present specification and claims and in the above figures, a number of operations are included that occur in a particular order, but it should be clearly understood that these operations may be performed out of order or in parallel as they occur herein, with the order of the operations being indicated as 101, 102, etc. merely to distinguish between the various operations, and the order of the operations by themselves does not represent any order of performance. Additionally, the flows may include more or fewer operations, and the operations may be performed sequentially or in parallel. It should be noted that, the descriptions of "first", "second", etc. in this document are used for distinguishing different messages, devices, modules, etc., and do not represent a sequential order, nor limit the types of "first" and "second" to be different.
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic flow chart of a postoperative END risk early warning method based on multimodal monitoring information according to an embodiment of the present invention, specifically, the method includes the following steps:
101: acquiring clinical data of a user to be detected in a current time period, wherein the clinical data comprises baseline data, dynamic monitoring data obtained in the current time period and dynamic monitoring data obtained in a historical time period;
in the embodiment of the invention, the baseline data comprises baseline information, past history information, baseline laboratory indexes, baseline image information, intraoperative conditions and other data.
In one embodiment, the patient inclusion criteria are: the age is more than or equal to 18 years old; patients with acute ischemic stroke; the time from disease onset to hospital is less than or equal to 24 hours; MRA or CTA or DSA suggests acute occlusion of the aorta; receiving intravascular treatment.
In one embodiment, clinical information such as baseline information, past history information, baseline laboratory indicators, baseline imaging information, intraoperative conditions, etc. includes basic information, demographic characteristics, pre-hospital first aid (thrombolytic venosus, intravascular treatment), past history, family history, past medication, admission subjects, admission diagnoses, baseline NIHSS score, auxiliary examinations during hospitalization, treatments during hospitalization, final diagnoses, drug taken out of hospital, NIHSS score out of hospital, time of occurrence, extent, cause of occurrence of END patients, and follow-up information; the imaging information includes completed image data within 24h of onset.
In one embodiment, the imaging information includes 1) preoperative images: baseline CT, multi-modality CT/MRI, etc.; 2) intraoperative imaging: DSA; 3) image review 24 hours after surgery: instant post-operative CT, 24-hour post-operative review MRI or CT.
In one embodiment, the baseline information (B) includes age, gender, baseline NIHSS score, baseline blood pressure, time to onset; the previous history information (H) comprises hypertension, coronary heart disease, atrial fibrillation and previous stroke, and the previous medication history (antiplatelet therapy and anticoagulant therapy); baseline laboratory indices (L) include admission blood glucose, white blood cell count, neutrophil count, PLT count, LDL); the baseline image information (I) comprises infarct volume, Mismatch volume, ASPECT score, infarct position, responsible blood vessel, occlusion degree, early signs, collateral circulation, leukoencephalopathy degree, microhemorrhage condition and the like; the intraoperative conditions (E) include intravenous thrombolysis treatment, anesthesia mode, intravascular treatment mode (stent removal, aspiration thrombolysis, stent formation, etc.), number of thrombolysis removal, intraoperative anticoagulation, antiplatelet application, postoperative TICI grading, residual stenosis, postoperative NIHSS score, time of surgery, etc.
In one embodiment, the basic monitoring parameters and the neuroelectrophysiological monitoring signals comprise basic vital sign monitoring information of 24 hours of postoperative continuous electrocardio, respiration, blood pressure, pulse oxygen and the like of the patients in the group; 24h electroencephalogram after operation and cerebral blood flow monitoring information. In some embodiments, the bedside multi-mode monitoring of basic monitoring parameters and neuroelectrophysiological monitoring signals shown in fig. 6 is used.
In one embodiment, the basic monitoring parameters and the neuroelectrophysiological monitoring signals include ECG (e), RESP (r), NIBP (nb), ABP (ab), SPO2(s), PULSE (p), EEG (eeg), TCD (tcd), etc.
In one embodiment, the clinical information is entered into the database online or offline through an electronic data capture system (EDC) based on a standardized design. The imaging data are stored in a DICOM format, and the imaging result interpretation adopts a centralized blind interpretation method. Intensive care record list, operation record, discharge summary, first page of medical record or out-patient emergency and other clinical information: and uploading the photo form. And (4) other checks: the cerebral blood flow and the electroencephalogram need to be stored in an EDF format or an ASCII format, and all information and data formats brought into a training set are qualified through background finger control.
In one embodiment, the basic care parameters include: ECG (electrocardiogram), RESP (respiration), NIBP (non-invasive blood pressure) or ABP (arterial blood pressure), SPO2 (blood oxygen), PULSE. The parameter requirements (including waveform, gain, filtering and the like) of the monitoring indexes are carried out according to the standard ICU monitoring mode requirements, bedside correction can be carried out on the condition of inaccurate signals, and the original data of all parameters are used for deep learning.
In one embodiment, bedside continuous EEG monitoring is performed on a patient using a mobile video electroencephalogram, requiring more than 24 hours of monitoring time. The electrodes were mounted 16 (except for cranial decompression patients) according to the international 10-20 system. The final electroencephalogram result analysis is performed by more than two professional interpreters and abnormal events (such as epileptiform discharges, NCS and the like) are calibrated, and the specific electroencephalogram interpretation is performed by adopting the American clinical neurophysiology institute intensive care electroencephalogram standard term.
In one embodiment, a bedside Transcranial Doppler ultrasound (TCD) examination is equipped with a 2.0MHz pulsed wave Doppler ultrasound probe using a Transcranial Doppler ultrasound machine. According to the TCD technical specification, the proper output power is formulated, and the sampling volume is selected to be 10-15 mm. And adjusting gain, a scale, a baseline and the like according to actual conditions. TCD examination at least includes MCA, ICA, V A, BA detection. Unless the window level is not good, each window level should be probed. All technical operations are carried out by a professional TCD technician, detection is carried out according to the content of a CRF form designed in advance, and finally result explanation is carried out by a technician with reading qualification.
In one embodiment, the clinical data is data-washed;
in one embodiment, data cleaning is mainly performed from two aspects, on one hand, for missing data processing, the data cleaning can be completed based on clinical analysis and model parameter optimization; on the other hand, the Binning method is adopted to treat the dispersion and the noise.
In one embodiment, the data washing step, the first step for missing data processing, can be done based on clinical analysis and model parameter optimization, the second step for dispersion and "noise" using Binning method, the third step for all data normalization processing.
102: inputting the clinical data into a depth self-encoder based on a convolution theory to obtain characteristic data corresponding to the clinical data;
acquiring a statistical value of the clinical data by using a statistical process control method;
in one embodiment, clinical data including high-dimensional clinical and image information and high-data-volume monitoring signals are subjected to feature extraction on postoperative patients without END and postoperative patients with END clinically through a depth automatic encoder based on convolution theory.
In one embodiment, in order to optimize the depth autoencoder structure based on convolution theory, a heuristic search algorithm is used to optimize the obtained encoding.
The specific flow of decoding and encoding is shown in the following equations (2-5).
x＝[B1,B2,B3..,H,H2,H3…,L1,L2,L3…,I,I2,I3…,E1,E2,E3…，e1，e2，e3…，r1,r2,r3…,nb,ab1,ab2,ab3…,s1,s2,s3…,p1,p2,p3…,eeg1,eeg2,eeg3…,tcd1,tcd2,tcd3…]T (1)
y＝f(x) (2)
x is a set of characteristics required by the model, wherein f (eta.) represents the coding process of x, and the code y is mapped and reconstructed to generate
g (.) represents the decoding process of y, and a sigmoid function (4) is adopted as a neural network activation function.
Equation (5) describes a method for obtaining optimal parameters during deep auto-encoder (DAE) training. Wherein WlAnd blIs the weight and offset of L layers, L is the number of DAE layers, and N is the number of samples in the data set. The internal structure of the depth encoder is shown in fig. 3.
A Deep Auto-encoder (DAE) is a Deep learning neural network structure which extracts the hierarchical features of high-dimensional complex input data from non-standard data by using a multi-layer nonlinear network with unsupervised pre-training and systematic parameter optimization and obtains the distributed feature representation of the original data. The DAE consists of an encoder, a decoder and an implicit layer. The encoder is a mapping of input y to an implicit representation h, represented as:
h＝f(x)＝St(W+bb) (6)
wherein S istA non-linear activation function, typically a logic function, whose expression is:
the decoder function g (h) maps the implicit layer data back to reconstruction y, as:
y＝g(h)＝Sg(W′h+by) (8)
wherein S isgIs the activation function of the decoder, typically a linear function or sigmoid function. The process of training the DAE is to find the parameter θ ═ W, b on the training sample set Dy,bhAnd (5) minimizing a reconstruction error, wherein the reconstruction error is expressed as:
JAE＝∑X∈DL(x，g(f(x))) (9)
wherein, L is a reconstruction error function, which can be generally expressed as a square error function or a cross entropy loss function, and the two are respectively expressed as:
L(x，y)＝||x-y||2 (10)
wherein the squared error is used for the linearity SgThe cross entropy loss function is used for sigmoid.
The DAE based on the convolution theory is a neural network which can be used for processing multi-modal data, the structure utilizes important local features to reconstruct original data, and all local features of input data share a weight matrix, so that the hidden layer of the DAE can completely save edge features limited by local space.
The implicit representation of the k-th order feature map for a single channel input x is:
hk＝σ(x*Wk+bk) (12)
wherein, σ is an activation function, a sigmoid function is generally adopted, "+" represents 2D convolution operation, and WkRepresenting a weight matrix, bkRepresenting the offset vector, the convolution operation expression is:
the reconstruction function for this class of DAE is:
where c is the bias for each data channel, H is the set of implicit feature maps,
the method is batch processing of the weight matrix, and the rule of updating the weight is random gradient descent. It is worth noting that in this class of DAE, one implicit mapping corresponds to one offset value, the offset vector b is valid for the entire mapping, and each mapping is responsible for capturing one feature of the data, which is convenient for pre-training and fine-engraving the neural network, effectively shortens the time of feature extraction, and simplifies featuresAnd the hierarchical extraction of the data features is realized in the process of feature extraction. The DAE based on the convolution theory is used for fitting the multi-modal data signal by using the linear combination of the basic modules, so that the speed and the accuracy of signal comprehensive identification are obviously improved. At present, the DAE can complete tasks such as target identification, dynamic following and simulation, and the problems of low identification speed, low accuracy, large amount of quasi-standard data and the like in the process of processing multi-modal monitoring data by the conventional DAE are effectively solved.
In one embodiment, selecting clinical data of a normal group of patients (END does not occur after operation) as a model training set, and inputting the clinical data of the training set into a depth self-encoder based on a convolution theory to obtain characteristic data corresponding to the clinical data; and acquiring input data and output data of the samples in each time period by using a statistical process control method, processing, and constructing a preliminary early warning model.
In one embodiment, the constructed preliminary early warning model is verified and improved by using an exponential weighted moving average control chart of statistical process control and clinical data of a verification set of patients (patients with END after operation and patients without END after operation) to obtain the early warning model. We use the clinical data relating to patients with END after surgery and define the parameter RE as the parameter for monitoring abnormalities, which is used to indicate the occurrence of END. Calculating an REs value (i is 1-N) through a trained depth encoder model, prompting that END has a larger probability when the value is larger, estimating a control limit of the END by using an exponential Weighted Moving Average control chart (EWMA) controlled by a statistical process, obtaining an early warning threshold value, and alarming (END early warning) when the model reaches the limit to prompt that the END will occur at a certain time in the future. The specific flow of feature extraction and model training of this segment is shown in fig. 4.
EMWA determines a predicted value by giving different weights to observed values and obtaining a moving average value by the different weights. The EMWA is adopted to better accord with the change characteristics of human biological signals, the method can not only combine the baseline data and the dynamic monitoring data of patients, but also meet the requirement that the recent observation value for observing the monitoring data has larger influence on the predicted value, and can reflect the trend of recent change. The exponentially weighted moving average control chart is defined by the formula:
Zi＝λxi+(1-λ)Zi-1 (15)
wherein the constant lambda is in the range of 0<λ≤1，ZiIs the EWMA statistic, i.e., the weighted average of all previous sample means.
Initial value Z of the formula0(when i is 1) taking the target value of the flow (i.e. at Z)0＝μ0) Sometimes, the mean value of the initial data is also used as the initial value, i.e. Z0XBar. Since EWMA is a weighted average of all previous and current samples, it is very insensitive to the normality assumption of the data and therefore ideal for a single observation.
If the observed values xi are independent random variables, the variance is σ2Then ZiHas a variance of
Thus, the ordinate axis of the EWMA control chart is ZiThe horizontal axis is sample number or time, and the calculation formula of the center line and the control limit is as follows:
Center line＝μ0
note that (1-lambda) in the above formula2iSection, when i is gradually increased, (1-lambda)2iWill soon converge to 0, so when i increases, UCL and LCL will settle to the following two values, which are the convergence of the control limit of EWMA:
in one embodiment, each time period can be any time period from 1 minute to 3 hours, and illustratively, each time period can be 1 minute, 2 minutes, 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes, 30 minutes, 1 hour, 1.5 hours, 2 hours, 2.5 hours, 3 hours, and the like.
Fig. 5 is a schematic diagram of the construction of the early warning model and the early warning provided in the embodiment of the present invention.
In one embodiment, the training mode of the depth self-encoder based on the convolution theory and the determination mode of the early warning threshold value of each time period comprise:
acquiring clinical data of a patient without END after operation in each time period as a training sample, and acquiring clinical data of the patient with END after operation in each time period as a verification sample;
training a depth self-encoder based on a convolution theory by using a training sample; and verifying the depth self-encoder based on the convolution theory by using the verification sample, processing the input data and the output data of the verification sample in each time period by using a statistical process control method to obtain an early warning threshold corresponding to each time period, and constructing an early warning model.
The method comprises the steps of taking clinical data of a user to be detected in a current time period, inputting the clinical data into a constructed early warning model, comparing a statistic value of the clinical data with an early warning threshold value corresponding to the current time period, and determining postoperative END risk early warning in the next time period according to a comparison result.
103: and comparing the statistical value of the clinical data with an early warning threshold value corresponding to the current time period, and determining postoperative END risk early warning of the next time period according to a comparison result.
In one embodiment, dynamic monitoring signals are continuously and automatically acquired through an intranet server 24 hours after operation, and are input into an early warning model, so that continuous real-time END early warning is realized.
Fig. 2 is a schematic block diagram of a post-operation END risk early warning device based on multi-modal monitoring information according to an embodiment of the present invention.
A postoperative END risk early warning device based on multi-modal monitoring information includes:
the system comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring clinical data of a user to be detected in a current time period, and the clinical data comprises baseline data, dynamic monitoring data obtained in the current time period and dynamic monitoring data obtained in a historical time period;
the processing unit is used for inputting the clinical data into a depth self-encoder based on a convolution theory to obtain characteristic data corresponding to the clinical data and acquiring a statistical value of the clinical data by using a statistical process control method;
and the prediction unit is used for comparing the statistic value of the clinical data with an early warning threshold value corresponding to the current time period and determining postoperative END risk early warning of the next time period according to a comparison result.
A postoperative END risk early warning method based on multi-modal monitoring information comprises the following steps:
acquiring clinical data of a user to be detected in a current time period, wherein the clinical data comprises baseline data, dynamic monitoring data obtained in the current time period and dynamic monitoring data obtained in a historical time period;
inputting the clinical data into a depth self-encoder based on a convolution theory to obtain characteristic data corresponding to the clinical data;
acquiring a statistical value of the clinical data by using a statistical process control method;
comparing the statistic value of the clinical data with an early warning threshold value corresponding to the current time period, and determining postoperative END risk early warning of the next time period according to the comparison result
A computer readable storage medium storing a computer program executed by a processor to implement the above-mentioned postoperative END risk early warning method based on multimodal monitoring information.
The validation results of this validation example show that assigning an intrinsic weight to an indication can moderately improve the performance of the method relative to the default settings.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by associated hardware instructed by a program, which may be stored in a computer-readable storage medium, and the storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.
It will be understood by those skilled in the art that all or part of the steps in the method for implementing the above embodiments may be implemented by hardware that is instructed to implement by a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
While the invention has been described in detail with reference to specific embodiments thereof, it will be apparent to one skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.
Claims (12)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110447993.7A CN113171059B (en) | 2021-04-25 | 2021-04-25 | Postoperative END risk early warning and related equipment based on multimodal monitoring information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110447993.7A CN113171059B (en) | 2021-04-25 | 2021-04-25 | Postoperative END risk early warning and related equipment based on multimodal monitoring information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113171059A CN113171059A (en) | 2021-07-27 |
| CN113171059B true CN113171059B (en) | 2022-03-01 |
Family
ID=76925245
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110447993.7A Active CN113171059B (en) | 2021-04-25 | 2021-04-25 | Postoperative END risk early warning and related equipment based on multimodal monitoring information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113171059B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115641916B (en) * | 2022-09-29 | 2024-05-03 | 杭州准芯生物技术有限公司 | Molecular biology detection method, computer equipment and readable storage medium |
| CN116719558B (en) * | 2023-08-09 | 2024-06-21 | 北京比瓴科技有限公司 | Gitlab baseline inspection method, gitlab baseline inspection device, gitlab baseline inspection equipment and readable storage medium |
| CN118039070B (en) * | 2024-04-11 | 2024-06-28 | 四川省肿瘤医院 | Clinical care system for interventional operation |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110556178A (en) * | 2018-05-30 | 2019-12-10 | 西门子医疗有限公司 | decision support system for medical therapy planning |
| CN110763997A (en) * | 2019-11-04 | 2020-02-07 | 华北电力大学（保定） | Early fault early warning method for synchronous motor stator |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10485486B2 (en) * | 2014-11-18 | 2019-11-26 | Baylor College Of Medicine | Clinical metric for predicting onset of cardiorespiratory deterioration in patients |
-
2021
- 2021-04-25 CN CN202110447993.7A patent/CN113171059B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110556178A (en) * | 2018-05-30 | 2019-12-10 | 西门子医疗有限公司 | decision support system for medical therapy planning |
| CN110763997A (en) * | 2019-11-04 | 2020-02-07 | 华北电力大学（保定） | Early fault early warning method for synchronous motor stator |
Non-Patent Citations (3)
| Title |
|---|
| 人工智能技术在神经重症监护中的应用;聂曦明 等;《临床荟萃》;20180405;第33卷(第4期);全文 * |
| 基于多模态传感数据的诊断预测方法与应用;王丹妮;《中国优秀硕士学位论文全文数据库》;20210115;第2021年卷(第01期);全文 * |
| 基于异常检测的血透患者死亡风险预测模型研究;娄国锋;《中国优秀硕士学位论文全文数据库》;20210215;第2021年卷(第02期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113171059A (en) | 2021-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113171059B (en) | Postoperative END risk early warning and related equipment based on multimodal monitoring information | |
| JP6857612B2 (en) | Cardiovascular deterioration warning score | |
| US8388530B2 (en) | Personalized monitoring and healthcare information management using physiological basis functions | |
| US20220093215A1 (en) | Discovering genomes to use in machine learning techniques | |
| EP1277433B1 (en) | Adjustable coefficients to customize predictive instruments | |
| RU2657384C2 (en) | Method and system for noninvasive screening physiological parameters and pathology | |
| Dai et al. | Intracranial pressure monitoring signals after traumatic brain injury: a narrative overview and conceptual data science framework | |
| KR102439082B1 (en) | A method of predicting the possibility of having liver disease in the general public using deep learning-based electrocardiogram analysis data | |
| CN115299887B (en) | Detection and quantification method and system for dynamic metabolic function | |
| CN116864104A (en) | Chronic thromboembolic pulmonary artery high-pressure risk classification system based on artificial intelligence | |
| Bodenes et al. | Early heart rate variability evaluation enables to predict ICU patients’ outcome | |
| Gupta et al. | A robust framework for automated screening of diabetic patient using ecg signals | |
| Panigrahy et al. | Predictive Modelling of Diabetes Complications: Insights from Binary Classifier on Chronic Diabetic Mellitus | |
| Panagiotou et al. | A multi: modal decision making system for an ambient assisted living environment | |
| TWI803893B (en) | Artificial intelligence assisted medical diagnosis method for sepsis and system thereof | |
| JP7563459B2 (en) | Analytical Equipment | |
| WO2023128863A2 (en) | Methods and systems for sepsis risk stratification | |
| Sahu et al. | Application of fuzzy logic to healthcare industry | |
| CN113208609A (en) | Electrocardio information management system | |
| Singh et al. | Classification of arrhythmia using transfer learning and its applications in smart wearables | |
| KR20210157444A (en) | Machine Learning-Based Diagnosis Method Of Schizophrenia And System there-of | |
| JP7562338B2 (en) | Computer-assisted method for non-invasive detection of carbohydrate metabolic disorders from electrocardiograms. | |
| JP7420266B2 (en) | Analysis equipment | |
| Bellos et al. | Categorization of COPD patient's health level through the use of the CHRONIOUS wearable platform | |
| US20250000445A1 (en) | Voice analysis to predict decompensated heart failure |
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 |