CN115019919B - Deep learning-based method for establishing function prediction model after early recovery of stroke - Google Patents

Deep learning-based method for establishing function prediction model after early recovery of stroke Download PDF

Info

Publication number
CN115019919B
CN115019919B CN202210681225.2A CN202210681225A CN115019919B CN 115019919 B CN115019919 B CN 115019919B CN 202210681225 A CN202210681225 A CN 202210681225A CN 115019919 B CN115019919 B CN 115019919B
Authority
CN
China
Prior art keywords
rehabilitation
information
time
model
mrs
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
Application number
CN202210681225.2A
Other languages
Chinese (zh)
Other versions
CN115019919A (en
Inventor
陆晓
郑瑜
张心彤
张秀
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangsu Province Hospital First Affiliated Hospital Of Nanjing Medical University
Original Assignee
Jiangsu Province Hospital First Affiliated Hospital Of Nanjing Medical University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jiangsu Province Hospital First Affiliated Hospital Of Nanjing Medical University filed Critical Jiangsu Province Hospital First Affiliated Hospital Of Nanjing Medical University
Priority to CN202210681225.2A priority Critical patent/CN115019919B/en
Publication of CN115019919A publication Critical patent/CN115019919A/en
Priority to PCT/CN2022/143730 priority patent/WO2023241012A1/en
Application granted granted Critical
Publication of CN115019919B publication Critical patent/CN115019919B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • 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
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • 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
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • 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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/20—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for electronic clinical trials or questionnaires
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • Pathology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Physiology (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Molecular Biology (AREA)
  • Veterinary Medicine (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Human Resources & Organizations (AREA)
  • Databases & Information Systems (AREA)
  • Marketing (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Game Theory and Decision Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Medicinal Chemistry (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)

Abstract

The invention provides a deep learning-based method for establishing a function prediction model after early stroke rehabilitation, which is characterized in that a mixed deep learning model consisting of a convolutional neural network and a long-short term memory artificial neural network is established, the early and accurate prediction is carried out on the function prognosis after the early stroke rehabilitation by combining clinical data and early rehabilitation related data, the long-term function prediction is carried out by combining a time-dependent resistance improved Rankin scale (mRS), and an individualized early rehabilitation strategy is guided; the machine learning-based functional prognosis prediction model after early rehabilitation of ischemic stroke, namely CNN-LSTM, can predict functions in early stage of stroke, provide accurate guidance for subsequent rehabilitation training scheme, better recover functions of stroke patients, save medical resources and reduce unnecessary consumption of manpower and material resources.

Description

Deep learning-based method for establishing function prediction model after early rehabilitation of stroke
Technical Field
The invention relates to a method for establishing a functional prognosis prediction model after early stroke rehabilitation, in particular to a method for establishing a functional prognosis prediction model after early stroke rehabilitation based on deep learning.
Background
The stroke has the characteristics of high morbidity, high mortality, high recurrence rate and the like, and according to the display of a global disease burden report 2010, the stroke becomes the second leading cause of death in the world and is also the disease with the highest disability rate of a single disease. Over the last two decades, treatment of patients with acute ischemic stroke has progressed tremendously with a significant reduction in mortality. However, as mortality decreases, the burden of disability increases in stroke survivors.
The treatment after acute ischemic stroke includes intravenous thrombolysis, mechanical embolectomy and the like, and the difficulty is how to evaluate the risk of a patient and how to obtain benefits from the treatment, thereby helping the decision of early treatment. Early rehabilitation is currently proposed as a means for promoting functional recovery of stroke patients and reducing mortality and disability rate. However, no research is available for predicting the functional prognosis level of acute ischemic stroke, and no research is available for predicting the functional and disability levels after early rehabilitation. However, early and accurate prediction of functional prognosis is of great reference for home decision-making for patients and family members. The prognosis of ischemic stroke is highly heterogeneous and difficult to predict. In recent years, there has been an increasing research into deep learning, which may provide a means to solve these challenging problems. Deep neural network analysis methods are adept at handling complex inputs and have been used to predict long-term prognosis.
The invention constructs a mixed deep learning model consisting of a convolutional neural network and a long-short term memory artificial neural network, performs early and accurate prediction on the function prognosis after early rehabilitation of ischemic stroke by combining clinical data and early rehabilitation related data, performs long-term function prediction by combining a time-dependent resistance improved Rankin scale (mRS), and guides and formulates an individualized early rehabilitation strategy.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provides a deep learning-based method for establishing a function prediction model after early stroke rehabilitation.
In order to achieve the purpose, the invention adopts the following technical scheme: a deep learning-based method for establishing a function prediction model after early rehabilitation of stroke comprises the following steps:
step S1: establishing a database of diseases;
preparing medical record data, collecting electronic medical records of patients from a hospital electronic medical record platform, and collecting electronic medical records of ischemic stroke patients for early rehabilitation; taking medical record data of a case with the first diagnosis of ischemic stroke as qualified electronic medical record data; step S2: extracting medical characteristic data of a patient;
extracting medical features of cerebral arterial thrombosis from the qualified electronic medical record data obtained in the step S1, and extracting medical features and medical feature values; the ischemic stroke characteristics comprise demographic information, laboratory and clinical examination related information, drug and invasive treatment related information, and rehabilitation intervention related information; material for use as a prediction;
the medical characteristic value is a specific numerical value of each medical characteristic in the characteristics of oral statistics, laboratory and clinical examination, medicine and invasive treatment and rehabilitation intervention;
the demographic information includes: gender, age, occupation, marital status, education, height, weight, BMI, systolic blood pressure, diastolic blood pressure, heart rate, whether it is a first cerebrovascular accident, TOAST typing, OCSP typing, past history, course of hypertension, course of diabetes, smoking status, age of cigarette, daily smoking count, smoking index, history of alcohol consumption, regular physical activity, family history; the laboratory and clinical examination related information includes: glycated hemoglobin, triglyceride, total cholesterol, low density lipoprotein cholesterol, high density lipoprotein cholesterol, lipoprotein a, homocysteine, partial thromboplastin time, prothrombin time-international normalized ratio, electrocardiogram, structural imaging examination results, common carotid artery stenosis, carotid bulbar stenosis, internal carotid artery stenosis, subclavian artery stenosis, left internal carotid intracranial stenosis, left anterior cerebral artery stenosis, left middle cerebral artery stenosis, left posterior cerebral artery stenosis, left vertebral artery stenosis, right internal carotid intracranial stenosis, right anterior cerebral artery stenosis, right middle cerebral artery stenosis, right posterior cerebral artery stenosis, right vertebral artery stenosis, basilar artery stenosis, swallowing function assessment, depression drinking test;
the drug and invasive treatment related information includes: intravenous thrombolysis, intravascular therapy, anti-platelet therapy within 48 hours, anticoagulant therapy within 48 hours, hypotensive drugs, lipid regulating drugs, hypoglycemic drugs;
the rehabilitation intervention related information comprises: the duration from disease onset to first rehabilitation intervention, the duration from disease onset to first mobilization, the benefit of early mobilization in first rehabilitation intervention, the duration of early mobilization in first rehabilitation intervention, the total duration of early mobilization in 14 days, the average duration of early mobilization in 14 days, the duration of physical therapy, the duration of operation therapy, the duration of speech therapy, continuous physical therapy for the first 14 days, continuous operation therapy for the first 14 days, and continuous speech therapy for the first 14 days;
and step S3: extracting target result characteristic data;
extracting stroke and scoring at different time steps, wherein the extracted time steps mainly comprise the day of hospital admission, 15 days after stroke, 30 days after stroke, 90 days after stroke and 180 days after stroke; extracting a post-stroke time score for predicting a target; time after stroke two classification results were: a favorable outcome is a time score of 0-2 after stroke, an unfavorable outcome is a time score of 3-6 after stroke, and the condition may be moderate or severe disability, or death;
big data information of ischemic stroke clinical manifestations can be obtained through the steps S1 to S3; and step S4: standardizing characteristic data and cleaning the data;
carrying out characteristic data standardization on the big data of the ischemic stroke clinical expression obtained in the step S3, adopting a missing data strategy, excluding patients with more than 50% missing of characteristic variables, filling the remaining characteristic missing data in a mode of existing data of the same characteristic, filling missing values of continuous variables by using a mean value, and filling missing values of classified variables by using a mode; all data are normalized to make the mean value and unit variance zero;
step S5: establishing a machine learning model 1-XGboost;
inputting the demographic information, laboratory and clinical examination related information, medicine and invasive treatment related information and rehabilitation intervention related information extracted in the step S2 into an XGboost model for mRS90 two-class prediction;
the XGboost comprises an XGboost decision tree and a relation between the XGboost decision trees; the XGboost decision tree comprises a plurality of nodes; the nodes are medical features and threshold values; the relation between the XGboost decision trees is a gradient descent optimization algorithm, and the next decision tree is obtained by the previous decision tree according to the gradient descent optimization algorithm;
and (3) feature screening in the XGboost model: automatically finding out the most relevant features by using XGboost, and using the most relevant features for mRS90 binary classification of a target result; training an estimator by utilizing initial characteristics on a development set, performing parameter adjustment or three-fold cross validation of hyper-parameter optimization by a grid search technology, generating ordered key characteristics by a trained model, and quantifying the relative importance of each variable by distributing a weight to each variable; the "weight" represents the total number of times the feature is used to partition data across all trees to measure feature importance in XGboost;
characteristic analysis in the XGboost model: calculating standard data samples, and screening relevant characteristics by using a statistical method such as T test, mann-Whitney U test and Kruskal-Wallis one-factor variance analysis; wherein T test, mann-Whitney U test, kruskal-Wallis one-factor analysis of variance are one of the methods commonly used in statistics; the statistical method and related software are used for calculation to obtain a probability value P, the value of P is set to be less than 0.05, the selected characteristics and the two classification targets of the ischemic mRS90 are considered to have extremely obvious correlation, and the characteristics are reasonably selected to establish a model; secondly, performing hierarchical clustering analysis on the screened characteristic variables and all rehabilitation intervention related information; the evaluation standard used by the hierarchical clustering is 'enclidean', and the method selects
Ward's method, the concrete realization adopts the open source tool library seaborn;
performing modeling experiments by using the selected demographic characteristic information and clinical characteristic information, all rehabilitation intervention related characteristic information and mRS as input information for the first time;
the modeling experiment refers to modeling in a development Set by using four machine learning algorithms of XGboost, SVM, random Forest (RF) and Logistic Regression (LR), in the modeling process, each machine learning method uses a grid searching method system to carry out automatic super-parameter optimization, in the grid searching process, F1score is used as a model evaluation standard, and 5-fold cross validation is adopted to select an optimal model;
step S6: establishing a machine learning model 2-CNN-LSTM;
the convolutional neural network-CNN is used as a backbone network to be combined with a long-short term memory network model-LSTM with a forgetting gate, and the rehabilitation of a patient at each time step is taken as a key point, and the time sequence modeling is carried out on the development condition of mRS rehabilitation.
The information adopted by the model comprises demographic characteristic information and clinical characteristic information screened out by XGboost, all characteristic information related to rehabilitation intervention, mRS (national standard language) grading and corresponding time step information; wherein the demographic characteristic information, the clinical characteristic information and all the characteristic information related to rehabilitation intervention belong to non-time sequence information, and the mRS score belongs to time sequence information; the mRS scores comprise scores of mRS-0, mRS-15, mRS-30, mRS-90 and mRS-180;
adopting the information as input information, constructing a network structure of a cascade convolutional neural network and a cyclic neural network, and in order to enable the non-time sequence state information of the patient to be obtained at each time step, firstly, using the convolutional neural network with stacked multilayer full-connection layers to perform feature aggregation and extraction on the non-time sequence state information, and finally using a sigmoid activation function as a score of the non-time sequence state information;
then, applying CNN to stack a plurality of complete connection layer aggregates, extracting characteristic discontinuous state information, and then generating non-time sequence state information by adopting the function;
combining the generated score with the time sequence information and the corresponding time step information, and fusing the generated score with the time sequence information and the corresponding time step information into an LSTM network;
training and learning the mRS rehabilitation development change of each patient by adopting an LSTM model;
finally, performing feature weighted fusion of the time steps by using an attention mechanism to enable the prediction of the mRS of each time step to be closer to the mRS of all time steps before the current time step;
step S7: establishing a machine learning model 3, stimulating observation and selecting key points;
testing and evaluating under different mRS missing conditions by using the trained CNN time sequence model so as to better explore the influence of mRS scoring at each time step in the follow-up process on the rehabilitation condition of the patient;
the CNN time sequence model is used for carrying out simulation modeling on rehabilitation progress by learning and developing the rehabilitation condition of each time step of a centralized patient, and the rehabilitation progress under different conditions is obtained by changing the input of the model;
comparing the influence degrees of the mRS scores under different time steps through the change of the model input; in the model, the final rehabilitation state of the patient is represented by mRS-180, and then the influence of the deletion of mRS score is explored and analyzed in three time steps of mRS-15, mRS-30 and mRS-90;
step S8: comparing results, and selecting a model;
and (4) comparing the results of the step (S5), the step (S6) and the step (S7) to obtain a CNN-LSTM model with the best specificity and sensitivity, and judging the CNN-LSTM model to be a prediction model of the functional prognosis after the early rehabilitation of the ischemic stroke.
Compared with the prior art, the invention has the following beneficial effects:
(1) The method adopts a machine learning CNN-LSTM integrated algorithm to establish a functional prognosis prediction model after early rehabilitation of the ischemic stroke, and accurately predicts the mRS-90 outcome of the ischemic stroke patient according to demographic information, laboratory and clinical examination related information, medicine and invasive treatment related information and rehabilitation intervention related information. The CNN-LSTM model is good in predicting the functional prognosis of early-stage rehabilitation ischemic stroke patients, and shows a prediction effect superior to that of four traditional algorithms. AUCs of the CNN-LSTM model in the test set were 0.829 (mRS-15), 0.706 (mRS-30), 0.809 (mRS-90) and 0.730 (mRS-180), respectively. In addition, the information of mRS-15 and mRS-30 is a key feature of CNN-LSTM model for prediction performance improvement of mRS-180.
(2) The functional prognosis prediction model CNN-LSTM based on machine learning for early rehabilitation of ischemic stroke can predict functions in early stroke, provide accurate guidance for subsequent rehabilitation training scheme, better recover functions of stroke patients, save medical resources and reduce unnecessary consumption of manpower and material resources.
Drawings
FIG. 1 is a flow chart of a method of model building according to the present invention.
Detailed Description
In order to further understand the objects, structures, features and functions of the present invention, the following embodiments are described in detail.
Referring to fig. 1, the invention provides a deep learning-based method for establishing a function prediction model after early recovery of stroke, which comprises the following steps:
step S1: establishing a database of diseases;
preparing medical record data, collecting electronic medical records of patients from a hospital electronic medical record platform, and collecting the electronic medical records of ischemic stroke patients who are recovered in early stage.
Step S2: extracting medical characteristic data of a patient;
extracting medical features of cerebral arterial thrombosis from the qualified electronic medical record data obtained in the step S1, and extracting medical features and medical feature values; the ischemic stroke characteristics include demographic information, laboratory and clinical examination related information, drug and invasive therapy related information, and rehabilitation intervention related information.
The medical characteristic values are specific numerical values for each of the medical characteristics of demographic, laboratory and clinical examinations, drugs and invasive treatments and rehabilitation interventions.
The demographic information includes: gender, age, occupation, marital status, education, height, weight, BMI, systolic blood pressure, diastolic blood pressure, heart rate, whether it is a first cerebrovascular accident, TOAST typing, OCSP typing, past history, course of hypertension, course of diabetes, smoking status, age of cigarette, daily smoking count, smoking index, history of alcohol consumption, regular physical activity, family history;
the history also includes ischemic stroke, hemorrhagic stroke, subarachnoid hemorrhage, unclassified stroke, hypertension, diabetes, dyslipidemia, atrial fibrillation, coronary heart disease, myocardial infarction, congenital heart disease, valvular heart disease, other types of heart diseases, peripheral arterial disease; the family history includes apoplexy, coronary heart disease, hypertension, diabetes, dyslipidemia and intracranial aneurysm.
The laboratory and clinical examination related information includes: glycated hemoglobin, triglyceride, total cholesterol, low density lipoprotein cholesterol, high density lipoprotein cholesterol, lipoprotein a, homocysteine, partial thromboplastin time, prothrombin time-international normalized ratio, electrocardiogram, structural imaging examination results, common carotid artery stenosis, carotid bulbar stenosis, internal carotid artery stenosis, subclavian artery stenosis, left internal carotid intracranial stenosis, left anterior cerebral artery stenosis, left middle cerebral artery stenosis, left posterior cerebral artery stenosis, left vertebral artery stenosis, right internal carotid intracranial stenosis, right anterior cerebral artery stenosis, right middle cerebral artery stenosis, right posterior cerebral artery stenosis, right vertebral artery stenosis, basilar artery stenosis, swallowing function assessment, depressed water drinking test.
The electrocardiogram detection comprises atrial fibrillation, atrial flutter, left ventricular hypertrophy, Q wave, acute myocardial infarction, myocardial ischemia and the like; the structural imaging examination results include hemorrhagic transformation after cerebral infarction, new cerebral infarction, old cerebral infarction and others.
The drug and invasive treatment related information includes: intravenous thrombolysis, intravascular therapy, anti-platelet therapy within 48 hours, anticoagulant therapy within 48 hours, antihypertensive drugs, lipid regulating drugs, and hypoglycemic drugs.
The intravascular treatment comprises stent thrombus removal, direct thrombus suction, balloon dilatation, intravascular stent-assisted angioplasty, intra-arterial thrombolysis and mechanical thrombolysis; said 48 hour anti-platelet therapy includes the use of aspirin, clopidogrel, ozagrel, dipyridamole, ticlopidine, cilostazol, and others; the anticoagulant therapy within 48 hours comprises warfarin, rivaroxaban, dabigatran, apixaban, edoxaban, low molecular heparin, plain heparin and others; the hypotensor comprises angiotensin converting enzyme inhibitor, angiotensin receptor blocker, diuretic, beta receptor blocker, calcium channel blocker and others; the lipid regulating drugs comprise statins, nicotinic acid and derivatives thereof, fibrates, cholesterol absorption inhibitors and the like; the hypoglycemic agent also comprises insulin, sulfonylurea, biguanide, glycosidase inhibitor, insulin sensitizer, insulin secretion promoter and other components.
The rehabilitation intervention related information comprises: the duration from disease onset to first rehabilitation intervention, the duration from disease onset to first mobilization, the benefit of early mobilization in first rehabilitation intervention, the duration of early mobilization in first rehabilitation intervention, the total duration of early mobilization in 14 days, the average duration of early mobilization in 14 days, the duration of physical therapy, the duration of operation therapy, the duration of speech therapy, continuous physical therapy for the first 14 days, continuous operation therapy for the first 14 days, and continuous speech therapy for the first 14 days.
And step S3: extracting target result characteristic data;
and (3) taking post-stroke time scores for predicting targets, and extracting mRS scores at different time steps, wherein the extraction time steps mainly comprise mRS-0 (the baseline is the hospital admission day), mRS-15 (15 days after stroke), mRS-30 (30 days after stroke), mRS-90 (90 days after stroke) and mRS-180 (180 days after stroke).
Time after stroke two classification results were: the beneficial result is a time score of 0-2 after stroke, which is manifested as no or minimal disability; adverse outcome post-stroke time scores 3-6, and the condition may be moderate or severe disability, or death. The large data of the clinical manifestations of ischemic stroke can be obtained through the steps S1 to S3.
And step S4: standardizing characteristic data and cleaning the data;
carrying out characteristic data standardization on the big data of the ischemic stroke clinical manifestation obtained in the step S3, adopting a missing data strategy, excluding patients with more than 50% missing of characteristic variables, filling the remaining characteristic missing data in a mode of the existing data with the same characteristic, filling missing values of continuous variables by using a mean value, and filling missing values of classified variables by using a mode; all data were normalized to zero mean and unit variance.
Step S5: establishing a machine learning model 1-XGboost;
inputting demographic information, laboratory and clinical examination related information, medicine and invasive treatment related information and rehabilitation intervention related information into an XGboost model for mRS90 two-class prediction;
the XGboost comprises an XGboost decision tree and a relation between the XGboost decision trees; the XGboost decision tree comprises a plurality of nodes; the nodes are medical features and threshold values; the relation between the XGboost decision trees is a gradient descent optimization algorithm, and the next decision tree is obtained by the previous decision tree according to the gradient descent optimization algorithm;
and (3) feature screening in the XGboost model: automatically finding out the most relevant features by using XGboost, and using the most relevant features for mRS90 binary classification of a target result; training an estimator by utilizing initial characteristics on a development set, performing parameter adjustment or three-fold cross validation of hyper-parameter optimization by a grid search technology, generating ordered key characteristics by a trained model, and quantifying the relative importance of each variable by distributing a weight to each variable; the "weight" represents the total number of times the feature is used to partition data across all trees to measure feature importance in XGboost;
characteristic analysis in the XGboost model: calculating standard data samples, and selecting relevant characteristics by a statistical method such as T test, mann-Whitney U test and Kruskal-Wallis single-factor analysis of variance; wherein T test, mann-Whitney U test, kruskal-Wallis one-factor analysis of variance are one of the methods commonly used in statistics; the statistical method and related software are used for calculation to obtain a probability value P, the value of P is set to be less than 0.05, the selected characteristics and the two classification targets of the ischemic mRS90 are considered to have extremely obvious correlation, and the characteristics are reasonably selected to establish a model; secondly, performing hierarchical clustering analysis on the screened characteristic variables and all rehabilitation intervention related information; the evaluation standard used by hierarchical clustering is ' includean ', and Ward's method is selected by the method, and the method specifically realizes the adoption of an open source tool library, namely, seaborn;
performing modeling experiments by using the selected demographic characteristic information and clinical characteristic information, all rehabilitation intervention related characteristic information and mRS as input information for the first time;
the modeling experiment refers to modeling in a development Set by using four machine learning algorithms of XGboost, SVM, random Forest (RF) and Logistic Regression (LR), wherein in the modeling process, each machine learning method uses a grid search method system to carry out automatic super-parameter tuning, and in the grid search process, F1score is used as a model evaluation standard, and an optimal model is selected by adopting 5-fold cross validation;
the feature information screened in the above steps includes the following:
demographic and clinical information: age of cigarette, hypoglycemic agent (biguanides), number of smoking cigarettes per day, past history (diabetes), antiplatelet therapy within 48 hours, anticoagulation therapy (others) within 48 hours, drinking history, family history (stroke), carotid artery angioscopy (carotid artery stenosis), family history (hypertension), occupation, diabetic course (years), family history (coronary heart disease), systolic blood pressure, whether or not first cerebrovascular accident, imaging (new cerebral infarction), education, swallowing function assessment, anticoagulation therapy within 48 hours, results of imaging examination (old cerebral infarction), lipoprotein (a), results of imaging examination (others), past history (hypertension), hypertensive course (years), hypotensive agent (angiotensin converting enzyme inhibitor), heart rate, gender, lipid regulating agent, age, triglyceride, OSCP typing, partial thromboplastin time, total cholesterol, body weight, protime-international normalized ratio, homocysteamine acid, high density lipoprotein sterol, glycated hemoglobin, cholesterol test, diastolic blood pressure, oral arterial stenosis, TOT intraarterial stenosis, right carotid stenosis, cervical lipoprotein density I, BMI; rehabilitation intervention related information: the duration (hours) from disease attack to first rehabilitation therapy, the duration (hours) from disease attack to first bed leaving, whether bed leaving is completed in first rehabilitation therapy, the bed leaving state maintaining time (minutes) in first bed leaving rehabilitation therapy, the total duration (minutes) of 14 days bed leaving, the average duration (minutes) of 14 days bed leaving, the duration (days) of physical therapy, the duration (days) of operation therapy, the duration (days) of speech therapy, 14 days of continuous physical therapy, 14 days of continuous operation therapy and 14 days of continuous speech therapy; time step information (day): taking values of 0, 15, 30, 90, 180.
Step S6: establishing a machine learning model 2-CNN-LSTM;
the convolutional neural network-CNN is used as a backbone network to be combined with a long-short term memory network model-LSTM with a forgetting gate, and the rehabilitation of a patient at each time step is taken as a key point, and the time sequence modeling is carried out on the development condition of mRS rehabilitation.
The information adopted by the model comprises demographic characteristic information and clinical characteristic information screened out by XGboost, all characteristic information related to rehabilitation intervention, mRS score and corresponding time step information; wherein the demographic characteristic information, the clinical characteristic information and all the characteristic information related to rehabilitation intervention belong to non-time sequence information, and the mRS score belongs to time sequence information; the mRS scores comprise scores of mRS-0, mRS-15, mRS-30, mRS-90 and mRS-180;
adopting the information as input information, constructing a network structure of a cascade convolution neural network and a cyclic neural network, and in order to enable the non-time sequence state information of the patient to be obtained at each time step, firstly, using the convolution neural networks of stacked multilayer full-connection layers to perform feature aggregation and extraction on the non-time sequence state information, and finally, using a sigmoid activation function as a score of the non-time sequence state information;
then, applying CNN to stack a plurality of complete connection layer aggregates, extracting characteristic discontinuous state information, and then generating non-time sequence state information by adopting the function;
combining the generated score with the time sequence information and the corresponding time step information, and fusing the score with the time sequence information and the corresponding time step information into an LSTM network;
training and learning the mRS rehabilitation development change of each patient by adopting an LSTM model;
finally, performing feature weighted fusion of the time steps by using an attention mechanism to enable the prediction of the mRS of each time step to be closer to the mRS of all time steps before the current time step;
the loss function used above is formulated as follows:
L total ＝θL mse +(1-θ)L fn
Figure GDA0004080016030000091
Figure GDA0004080016030000092
wherein L is total Denotes the overall loss, L mse Represents a weighted mean square error loss function, L fn Represents a weighted local loss function (which can effectively reduce the problem of complicated category imbalance among a plurality of time steps and is difficult to sample and mine), mse represents a mean square error loss function (which can make the prediction effectiveness of different time steps more balanced), and W represents a mean square error loss function (which can make the prediction effectiveness of different time steps more balanced) mask Represents a weighting coefficient, theta represents a weighting factor, and the value range is [0,1 ]]The value of the model is 0.25; p represents a prediction result, P' represents a label, N represents the number of samples, alpha is a weighing failure, and the value range is [0,1 ]]The value of the model is 1; gamma represents the focusing parameter, and the value range is generally [0,5']The value of the model is 2.
Step S7: establishing a machine learning model 3, stimulating observation and selecting key points;
testing and evaluating under different mRS missing conditions by using a trained time sequence model so as to better explore the influence of mRS scoring at each time step in the follow-up process on the recovery condition of the patient;
the time sequence model is used for centralizing the rehabilitation condition of each time step of the patient through learning and development, carrying out simulation modeling on the rehabilitation progress, and changing the input of the model to obtain the rehabilitation progress under different conditions;
comparing the influence degrees of the mRS scores under different time steps through the change of the model input; in this model, the patient's final recovery status is represented by mRS-180, and then the effects of the absence of mRS scores are analyzed using three time steps mRS-15, mRS-30, and mRS-90.
Step S8: comparing results, and selecting a model;
and (4) comparing the results of the step (S5), the step (S6) and the step (S7) to obtain a CNN-LSTM model with the best specificity and sensitivity, and judging the CNN-LSTM model to be a prediction model of the functional prognosis after the early rehabilitation of the ischemic stroke.
The present invention has been described in relation to the above embodiments, which are only examples of the implementation of the present invention. It should be noted that the disclosed embodiments do not limit the scope of the invention. Rather, it is intended that all such modifications and variations be included within the spirit and scope of this invention.

Claims (4)

1. A deep learning-based method for establishing a function prediction model after early rehabilitation of stroke is characterized by comprising the following steps: the model establishing method comprises the following steps:
step S1: establishing a database of diseases;
preparing medical record data, collecting electronic medical records of patients from a hospital electronic medical record platform, and collecting the electronic medical records of ischemic stroke patients who are recovered in early stage; taking medical record data of a case with the first diagnosis of ischemic stroke as qualified electronic medical record data;
step S2: extracting medical characteristic data of a patient;
extracting medical features of cerebral arterial thrombosis from the qualified electronic medical record data obtained in the step S1, and extracting medical features and medical feature values; the ischemic stroke characteristics comprise demographic information, laboratory and clinical examination related information, drug and invasive treatment related information, and rehabilitation intervention related information; material for use as a prediction;
and step S3: extracting target result characteristic data;
extracting stroke and scoring at different time steps, wherein the extracted time steps comprise the day of hospital admission, 15 days after stroke, 30 days after stroke, 90 days after stroke and 180 days after stroke; a post-stroke time score for the prediction target is extracted,
time after stroke two classification results were: favorable outcome is 0-2 post-stroke time score, unfavorable outcome is 3-6 post-stroke time score, and condition includes moderate or severe disability, or death;
big data information of ischemic stroke clinical manifestations can be obtained through the steps S1 to S3;
and step S4: standardizing characteristic data and cleaning the data;
carrying out characteristic data standardization on the big data of the ischemic stroke clinical expression obtained in the step S3, adopting a missing data strategy, excluding patients with more than 50% missing of characteristic variables, filling the remaining characteristic missing data in a mode of existing data of the same characteristic, filling missing values of continuous variables by using a mean value, and filling missing values of classified variables by using a mode; all data are normalized to make the mean value and unit variance zero;
step S5: establishing a machine learning model 1-XGboost;
inputting the demographic information, laboratory and clinical examination related information, medicine and invasive treatment related information and rehabilitation intervention related information extracted in the step S2 into an XGboost model for mRS90 two-class prediction;
the XGboost model comprises an XGboost decision tree and a relation between the XGboost decision trees; the XGboost decision tree comprises a plurality of nodes; the nodes are medical features and threshold values; the relation between the XGboost decision trees is a gradient descent optimization algorithm, and the next decision tree is obtained by the previous decision tree according to the gradient descent optimization algorithm; carrying out feature screening and feature analysis on the XGboost model; characteristic variables are screened out and modeled by using an XGboost mode, and an XGboost prediction model is finally established;
step S6: establishing a machine learning model 2-CNN-LSTM;
the convolutional neural network CNN is used as a backbone network to be combined with a long-short term memory network model-LSTM with a forgetting gate, and the time sequence modeling is carried out by taking the rehabilitation of a patient at each time step as a key point and the development condition of mRS rehabilitation;
the information adopted by the CNN-LSTM model comprises demographic characteristic information and clinical characteristic information screened out by XGboost, all characteristic information related to rehabilitation intervention, mRS score and corresponding time step information; wherein the demographic characteristic information, the clinical characteristic information and all the characteristic information related to rehabilitation intervention belong to non-time sequence information, and the mRS score belongs to time sequence information; the mRS score comprises scores of mRS-0, mRS-15, mRS-30, mRS-90 and mRS-180;
the demographic characteristic information, the clinical characteristic information, all rehabilitation intervention related characteristic information, mRS scores and corresponding time step information are used as input information to construct a network structure of a cascade convolutional neural network and a cyclic neural network, in order to enable each time step to obtain non-time sequence state information of a patient, feature aggregation and extraction are firstly carried out on the non-time sequence state information by using the convolutional neural network with stacked layers and full connection layers, and finally a sigmoid activation function is used as a score of the non-time sequence state information; then, a CNN is applied to stack a plurality of complete connection layer aggregates, state information with discontinuous characteristics is extracted, and then the sigmoid activation function is adopted to generate non-time sequence state information; combining the generated score with the time sequence information and the corresponding time step information, and fusing the score with the time sequence information and the corresponding time step information into an LSTM network; training and learning the mRS rehabilitation development change of each patient by adopting an LSTM model; finally, performing feature weighted fusion of the time steps by using an attention mechanism to enable the prediction of the mRS of each time step to be closer to the mRS of all time steps before the current time step;
s6, establishing a CNN-LSTM prediction model by utilizing the characteristic variables screened in the step S5 and using a convolution neural network-CNN as a mode of combining a main network with a long-short term memory network model with a forgetting gate-LSTM;
step S7: establishing a machine learning model 3, stimulating observation and selecting key points;
testing and evaluating under different mRS deficiency conditions by using the CNN-LSTM model trained in the step S6 so as to better explore the influence of mRS scores at each time step in the follow-up process on the rehabilitation condition of the patient;
the CNN-LSTM model is used for carrying out simulation modeling on rehabilitation progress by centralizing rehabilitation conditions of each time step of a patient through learning and development, and changing the input of the model to obtain the rehabilitation progress under different conditions;
comparing the influence degree of the mRS score under different time steps through the change of the CNN-LSTM model input; in the model, the final rehabilitation state of the patient is represented by mRS-180, and then the influence of the deletion of mRS score is explored and analyzed in three time steps of mRS-15, mRS-30 and mRS-90;
step S7 is to repeat the modeling process of the step S6 for a plurality of times, but one variable is reduced on the basis of the step S6 when modeling is performed each time, namely the mRS score is reduced at one time step each time, and different modeling results are compared to obtain a model with the best prediction efficiency;
step S8: comparing results, and selecting a model;
and (4) comparing the results of the step (S5), the step (S6) and the step (S7) to obtain a CNN-LSTM model with the best specificity and sensitivity, and judging the CNN-LSTM model to be a prediction model of the functional prognosis after the early rehabilitation of the ischemic stroke.
2. The deep learning-based method for establishing the function prediction model after early rehabilitation of stroke as claimed in claim 1, wherein: in the step S2, the medical characteristic value is a specific numerical value of each medical characteristic in the characteristics of oral statistics, laboratory and clinical examinations, drugs and invasive treatments and rehabilitation interventions; the demographic information includes gender, age, occupation, marital status, education, height, weight, BMI, systolic blood pressure, diastolic blood pressure, heart rate, whether it is a first cerebrovascular accident, TOAST typing, OCSP typing, past history, course of hypertension, course of diabetes, smoking status, tobacco age, daily smoking count, smoking index, drinking history, regular physical activity, and family history; information related to the laboratory and clinical examinations includes glycated hemoglobin, triglyceride, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, lipoprotein a, homocysteine, partial thromboplastin time, prothrombin time-international standardized ratio, electrocardiogram, structural imaging examination results, common carotid artery stenosis, carotid bulbar stenosis, internal carotid artery stenosis, subclavian artery stenosis, left internal carotid intracranial stenosis, left anterior cerebral artery stenosis, left middle cerebral artery stenosis, left posterior cerebral artery stenosis, left vertebral artery stenosis, right internal carotid intracranial stenosis, right anterior cerebral artery stenosis, right middle cerebral artery stenosis, right posterior cerebral artery stenosis, right vertebral artery stenosis, basilar artery stenosis, swallowing function assessment, and water drinking test in a hollow field; the drug and invasive treatment related information includes: intravenous thrombolysis, intravascular treatment, antiplatelet treatment within 48 hours, anticoagulation treatment within 48 hours, antihypertensive drugs, lipid-regulating drugs, hypoglycemic drugs; the rehabilitation intervention related information comprises the time length from disease onset to first rehabilitation intervention, the time length from disease onset to first mobilization, the benefit of early mobilization in first rehabilitation intervention, the early mobilization time length in first rehabilitation intervention, the total time length of early mobilization in 14 days, the average time length of early mobilization in 14 days, the physical therapy time length, the operation therapy time length, the speech therapy time length, continuous physical therapy for 14 days in the first time, continuous operation therapy for 14 days in the first time, and continuous speech therapy for 14 days in the first time.
3. The deep learning-based method for establishing the function prediction model after early stroke rehabilitation as claimed in claim 1, wherein: the XGboost model feature screening in the step S5 is to automatically find out the most relevant features by using XGboost, and the most relevant features are used for mRS90 binary classification of target results; training an estimator by utilizing initial characteristics on a development set, performing parameter adjustment or three-fold cross validation of hyper-parameter optimization by a grid search technology, generating ordered key characteristics by a trained model, and quantifying the relative importance of each variable by distributing a weight to each variable; the XGboost model feature analysis refers to the calculation of standard data samples, and the statistical method for screening relevant features is T test, mann-WhitneyU test and Kruskal-Wallis single-factor analysis of variance; secondly, performing hierarchical clustering analysis on the screened characteristic variables and all rehabilitation intervention related information; the evaluation standard used by the hierarchical clustering is ' enclidean ', the method selects Ward ' small, and an open source tool library, namely, seaborn, is specifically adopted; modeling experiments were performed using the selected demographic and clinical profile information, all rehabilitation intervention related profile information, and mRS as input information for the first time.
4. The deep learning-based method for establishing the function prediction model after early stroke rehabilitation as claimed in claim 3, wherein: the modeling experiment refers to modeling in a development set by using four machine learning algorithms of XGboost, SVM, randomforest (RF) and Logistic Regression (LR), wherein in the modeling process, each machine learning method uses a grid searching method system to carry out automatic super-parameter optimization, and in the grid searching process, F1score is used as a model evaluation standard and an optimal model selected by 5-fold cross validation is adopted.
CN202210681225.2A 2022-06-16 2022-06-16 Deep learning-based method for establishing function prediction model after early recovery of stroke Active CN115019919B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202210681225.2A CN115019919B (en) 2022-06-16 2022-06-16 Deep learning-based method for establishing function prediction model after early recovery of stroke
PCT/CN2022/143730 WO2023241012A1 (en) 2022-06-16 2022-12-30 Method for establishing deep learning-based model for predicting functions after post-cerebral stroke early rehabilitation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210681225.2A CN115019919B (en) 2022-06-16 2022-06-16 Deep learning-based method for establishing function prediction model after early recovery of stroke

Publications (2)

Publication Number Publication Date
CN115019919A CN115019919A (en) 2022-09-06
CN115019919B true CN115019919B (en) 2023-04-07

Family

ID=83074245

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210681225.2A Active CN115019919B (en) 2022-06-16 2022-06-16 Deep learning-based method for establishing function prediction model after early recovery of stroke

Country Status (2)

Country Link
CN (1) CN115019919B (en)
WO (1) WO2023241012A1 (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115019919B (en) * 2022-06-16 2023-04-07 江苏省人民医院（南京医科大学第一附属医院） Deep learning-based method for establishing function prediction model after early recovery of stroke
CN116403714B (en) * 2023-04-07 2024-01-26 大连市中心医院 Cerebral apoplexy END risk prediction model building method and device, END risk prediction system, electronic equipment and medium
CN117524464B (en) * 2024-01-04 2024-04-05 北京和兴创联健康科技有限公司 Method and system for calculating postoperative target hemoglobin based on big data
CN117637154B (en) * 2024-01-27 2024-03-29 南通大学附属医院 Nerve internal department severe index prediction method and system based on optimization algorithm
CN117828490B (en) * 2024-03-06 2024-05-17 南京信息工程大学 Typhoon disaster forecasting method and system based on ensemble learning
CN117995410B (en) * 2024-03-08 2024-06-28 南京医科大学 Method and device for male sterility risk prediction and risk feature description

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113040711A (en) * 2021-03-03 2021-06-29 吾征智能技术(北京)有限公司 Cerebral stroke attack risk prediction system, equipment and storage medium
CN114587385A (en) * 2022-02-25 2022-06-07 之江实验室 Method for constructing post-stroke rehabilitation assessment deep learning model based on brain muscle network graph theory characteristics

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101740464B1 (en) * 2016-10-20 2017-06-08 (주)제이엘케이인스펙션 Method and system for diagnosis and prognosis of stroke and systme therefor
KR102190299B1 (en) * 2017-02-02 2020-12-11 사회복지법인 삼성생명공익재단 Method, device and program for predicting the prognosis of gastric cancer using artificial neural networks
CN111513704B (en) * 2020-06-09 2021-06-25 首都医科大学附属北京天坛医院 Index and system for predicting prognosis of acute ischemic stroke patient receiving revascularization treatment
CN111899894B (en) * 2020-08-03 2021-06-25 东南大学 System and method for evaluating prognosis drug effect of depression patient
CN113808747B (en) * 2021-10-11 2023-12-26 南昌大学第二附属医院 Ischemic cerebral apoplexy recurrence prediction method
CN113988209A (en) * 2021-11-09 2022-01-28 昆明同心医联科技有限公司 Training method and device of stroke prognosis prediction model
CN114203295B (en) * 2021-11-23 2022-05-20 国家康复辅具研究中心 Cerebral apoplexy risk prediction intervention method and system
CN115019919B (en) * 2022-06-16 2023-04-07 江苏省人民医院（南京医科大学第一附属医院） Deep learning-based method for establishing function prediction model after early recovery of stroke

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113040711A (en) * 2021-03-03 2021-06-29 吾征智能技术(北京)有限公司 Cerebral stroke attack risk prediction system, equipment and storage medium
CN114587385A (en) * 2022-02-25 2022-06-07 之江实验室 Method for constructing post-stroke rehabilitation assessment deep learning model based on brain muscle network graph theory characteristics

Also Published As

Publication number Publication date
WO2023241012A1 (en) 2023-12-21
CN115019919A (en) 2022-09-06

Similar Documents

Publication Publication Date Title
CN115019919B (en) Deep learning-based method for establishing function prediction model after early recovery of stroke
Sk et al. Coronary Heart Disease Prediction and Classification using Hybrid Machine Learning Algorithms
Rajendra et al. Prediction of diabetes using logistic regression and ensemble techniques
US11735321B2 (en) System for the prognostics of the chronic diseases after the medical examination based on the multi-label learning
CN109920547A (en) A kind of diabetes prediction model construction method based on electronic health record data mining
CN109273093A (en) A kind of construction method and building system of Kawasaki disease risk evaluation model
Altikardes et al. A novel method for dipper/non-dipper pattern classification in hypertensive and non-diabetic patients
CN113128654B (en) Improved random forest model for coronary heart disease pre-diagnosis and pre-diagnosis system thereof
Mansur Huang et al. Machine learning techniques for early heart failure prediction
US20210089965A1 (en) Data Conversion/Symptom Scoring
Zhou et al. Prediction of 1-year mortality after heart transplantation using machine learning approaches: A single-center study from China
CN114820490A (en) Aggregated platelet detection method based on YOLOX improved algorithm
Adi et al. Stroke risk prediction model using machine learning
Ghffar et al. Usefulness of semisupervised machine-learning-based phenogrouping to improve risk assessment for patients undergoing transcatheter aortic valve implantation
CN116580847B (en) Method and system for predicting prognosis of septic shock
Wadhawan et al. ETCD: An effective machine learning based technique for cardiac disease prediction with optimal feature subset selection
Tahir et al. Neural networks algorithm to inquire previous preeclampsia factors in women with chronic hypertension during pregnancy in childbirth process
Padmapritha Prediction of Blood Glucose Level by using an LSTM based Recurrent Neural networks
CN117493970A (en) Deep learning-based prediction method for adverse cardiovascular events after PCI (peripheral component interconnect) operation of coronary heart disease patient
Vilas-Boas et al. Hourly prediction of organ failure and outcome in intensive care based on data mining techniques
CN114864097A (en) Method and device for establishing aorta dissection patient postoperative death prediction model
CN113782197B (en) New coronary pneumonia patient outcome prediction method based on interpretable machine learning algorithm
Mondal et al. Early prediction of coronary heart disease using boosting-based voting ensemble learning
Riyaz et al. Ensemble learning for coronary heart disease prediction
Chao et al. Topological data analysis identified prognostically-distinct phenotypes in transcatheter edge-to-edge repair patients

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