A method for assessing lower limb prognosis in stroke patients for brain-computer interface rehabilitation
By extracting and processing multimodal clinical medical data from stroke patients, using elastic network and artificial neural network models to evaluate their responsiveness to BCI rehabilitation therapy, the problem of lack of prognosis evaluation of stroke patients in the prior art is solved, and personalized The auxiliary treatment improves the efficiency of BCI rehabilitation therapy.
Patent Information
- Application Number
- CN202410360949.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-27
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-03-27
AI Technical Summary
In the prior art, there is a lack of a method for evaluation of the prognosis of stroke patients with brain-computer interface rehabilitation, and it is impossible to effectively assist in the selection of personalized therapies to maximize the rehabilitation efficiency of BCI therapy.
Multimodal clinical medical data, including demographics, clinical scales, and biomechanical characteristics, were extracted, and prognostic evaluation was performed through elastic network models and artificial neural network models to determine the optimal model to judge the responsiveness of stroke patients to BCI rehabilitation therapy.
Through standardized processing and model evaluation of multimodal data, the proportion of lower limb motor function recovery in stroke patients can be accurately judged, and auxiliary physicians can customize personalized therapy to improve the efficiency of BCI rehabilitation therapy.
Smart Images

Figure CN118173268B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of clinical diagnosis and medicine, and in particular to a method for assessing the prognosis of lower limbs in stroke patients for brain-computer interface rehabilitation. Background Art
[0002] In the prior art, stroke, also known as infarction, is an acute cerebrovascular disease that causes brain tissue damage due to sudden rupture of brain blood vessels or blockage of blood vessels that prevent blood from flowing into the brain. According to statistics from the health department, the total number of stroke patients in my country has approached 20 million, and the number of new patients exceeds 2.5 million each year. Data show that in 2019, the incidence of ischemic stroke in my country was 1,700 / 100,000 and the incidence of hemorrhagic stroke was 306 / 100,000. Stroke is also the main cause of disability and is a huge economic burden on patients and their families.
[0003] Non-invasive brain-computer interfaces (BCI) can transform endogenous neural activity into meaningful outputs, thereby promoting state-dependent neuroplasticity and functional corticospinal connectivity. A systematic meta-analysis reported that BCI-based neurorehabilitation can improve the outcome of motor injuries more than traditional treatments and induce functional and structural neuroplasticity at a subclinical level. BCI rehabilitation may be a promising rehabilitation therapy, but due to the high individual variability of motor recovery after stroke, the "one-size-fits-all" strategy of relying solely on a single therapy is limited in clinical practice. Some studies have also shown that there are patients who respond to BCI therapy and patients who do not respond, that is, patients who recover better and patients who recover poorly. Therefore, early and accurate prognostic assessment of stroke patients for BCI rehabilitation is very necessary, which can assist in selecting personalized therapies suitable for patient characteristics, customize treatments, and maximize the rehabilitation efficiency of BCI therapy.
[0004] With the continuous development of the field of stroke rehabilitation, the prognosis evaluation of rehabilitation treatment has received more and more attention. In recent years, several studies have used biological metabolomics, clinical data, blood flow in specific brain regions and other characteristics to predict the functional recovery level of patients and establish prognosis evaluation models, such as patent CN115019919A (China, 2022.09.06), which constructs a hybrid deep learning model consisting of a convolutional neural network and a long short-term memory artificial neural network, combined with clinical data and early rehabilitation-related data to make early and accurate predictions of functional prognosis after early rehabilitation of ischemic stroke. However, most of the current studies focus on the prognosis evaluation problem in the context of traditional therapy rehabilitation, and do not consider the prognosis evaluation of stroke patients for BCI rehabilitation. Summary of the invention
[0005] The present application aims to solve one of the technical problems in the related art at least to some extent.
[0006] To this end, the present application embodiment proposes a method for assessing the prognosis of lower limbs of stroke patients for brain-computer interface rehabilitation, including:
[0007] Extracting multimodal clinical medical data of stroke patients for BCI rehabilitation, wherein the multimodal clinical medical data includes demographic characteristics, clinical scale characteristics, and biomechanical characteristics;
[0008] Calculating the recovery rate of stroke patients according to the multimodal clinical medical data, and assigning labels to stroke patients with different recovery rates;
[0009] Standardizing the multimodal clinical medical data, and inputting the standardized multimodal clinical medical data into a trained elastic network model Elastic net and an artificial neural network model ANN, respectively, to obtain a first prognostic evaluation result and a second prognostic evaluation result;
[0010] Based on the actual results after the label division, the prediction capabilities of Elastic Net and ANN are compared according to the first prognostic evaluation results and the second prognostic evaluation results to determine the optimal model;
[0011] Based on the optimal model, the prognosis of stroke patients undergoing BCI rehabilitation is evaluated.
[0012] Optionally, the extracting multimodal clinical medical data of stroke patients includes:
[0013] Based on the hospital's electronic medical record platform, the electronic medical records of subacute stroke patients who received BCI rehabilitation training after admission were collected to establish a patient information database;
[0014] The multimodal clinical medical data of stroke patients is extracted from the patient information database.
[0015] Optionally, the demographic characteristics include age, gender, stroke type and disease duration;
[0016] The clinical scale characteristics include motor ability, balance function, walking ability, daily living ability, and spasticity level;
[0017] The biomechanical characteristics include muscle strength level and horizontal range of motion.
[0018] Optionally, calculating the recovery ratio of the stroke patients according to the multimodal clinical medical data and assigning labels to stroke patients with different recovery ratios includes:
[0019] According to the Fugl-Meyer lower limb motor function score FMA-LE before and after BCI rehabilitation training, the recovery rate of lower limb motor function in stroke patients was calculated, and the expression was:
[0020]
[0021] Among them, FMA-LE pre 、FMA-LE post refers to the FMA-LE scores before and after BCI rehabilitation training, respectively, maximumscore refers to the total score of the FMA-LE scale, and Recovery Rroportion refers to the recovery proportion;
[0022] The median recovery rate of all stroke patient samples included was calculated, and stroke patients with a recovery rate higher than the median were classified as responders, while stroke patients with a recovery rate lower than the median were classified as non-responders.
[0023] Optionally, the standardization processing of the multimodal clinical medical data includes:
[0024] The multimodal clinical medical data is standardized using the z-score standardization method to make its mean zero and variance one, and the expression is:
[0025]
[0026] Among them, x′ refers to the standardized features, x refers to the original features, mean refers to the average value of the original features of all included samples, and sd refers to the standard deviation of the original features of all included samples.
[0027] Optional, the training process of the elastic network model Elastic net includes:
[0028] The leave-one-out cross-validation strategy is adopted to generate several different training sets and test sets from multimodal clinical medical data samples;
[0029] In each split, a second round of leave-one-out cross-validation is performed using a grid search method to search for the optimal set of hyperparameters that produces the highest average classification accuracy in leave-one-out cross-validation to optimize the regularization parameter λ and the mixing parameter ρ of the Elastic net model in the training set.
[0030] The optimal hyperparameters are set on the entire training set to train the Elasticnet model and predict the category of the test set.
[0031] Optionally, the training process of the artificial neural network model ANN includes:
[0032] The leave-one-out cross-validation strategy is adopted to generate several different training sets and test sets from multimodal clinical medical data samples;
[0033] A second round of leave-one-out cross-validation was performed using a grid search method in each split to search for the optimal hyperparameter set that produced the highest average classification accuracy in the leave-one-out cross-validation to optimize the number of hidden layer nodes m and the L2 regularization parameter α of the ANN model in the training set;
[0034] The optimal hyperparameters are set on the entire training set to train the ANN model and predict the category of the test set.
[0035] Optionally, the process of inputting the standardized multimodal clinical medical data into a trained Elastic net for prediction includes:
[0036] Inputting the standardized multimodal clinical medical data into the trained Elastic net model to perform binary classification prediction of recovery ratio;
[0037] Based on the Elastic net model, L1 and L2 regularization were applied simultaneously to predict the first FMA-LE score after BCI rehabilitation training through linear regression;
[0038] According to the first FMA-LE score, the first predicted recovery ratio of the lower limb motor function of the stroke patient is calculated, and the stroke patients with different first predicted recovery ratios are labeled according to the first predicted recovery ratio to obtain a first prognostic assessment result.
[0039] Optionally, the process of inputting the standardized multimodal clinical medical data into a trained ANN for prediction includes:
[0040] Inputting the standardized multimodal clinical medical data into the trained ANN model to perform binary classification prediction of recovery ratio;
[0041] Based on the ANN model, the features in the input multimodal clinical medical data are aggregated and extracted, and the second FMA-LE score after BCI rehabilitation training is predicted through a sigmoid activation function;
[0042] According to the second FMA-LE score, the second predicted recovery ratio of the lower limb motor function of the stroke patients is calculated, and the stroke patients with different second predicted recovery ratios are labeled according to the second predicted recovery ratio to obtain a second prognostic assessment result.
[0043] Optionally, the actual result after the label division is compared with the prediction ability of Elastic Net and ANN according to the first prognosis evaluation result and the second prognosis evaluation result to determine the optimal model, including:
[0044] The true result after the label division and the first prognosis evaluation result are evaluated according to evaluation indicators to obtain a first evaluation result, wherein the evaluation indicators include classification accuracy, area under the curve, specificity and sensitivity;
[0045] The true result after the label division and the second prognosis evaluation result are evaluated according to the evaluation index to obtain a second evaluation result;
[0046] If the first evaluation result is better than the second evaluation result, the Elastic net model is determined to be the optimal model; otherwise, the ANN is determined to be the optimal model.
[0047] The technical solution provided by the embodiments of the present application brings at least the following beneficial effects:
[0048] By extracting multimodal clinical medical data of stroke patients undergoing BCI rehabilitation and inputting them into different models for evaluation to obtain the optimal model, the responsiveness to BCI rehabilitation therapy can be judged based on the patient's demographics, clinical scales, and biomechanical characteristics, thereby assisting physicians in customizing personalized treatments for patients.
[0049] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:
[0051] Figure 1 It is a flow chart of a method for evaluating the prognosis of lower limbs of stroke patients for brain-computer interface rehabilitation according to an embodiment of the present application;
[0052] Figure 2 are input features and prediction features for constructing a prediction model according to an embodiment of the present application;
[0053] Figure 3 is a confusion matrix result diagram of model comparison according to an embodiment of the present application;
[0054] Figure 4 It is a ROC curve result diagram of the model comparison shown in the embodiment of the present application. DETAILED DESCRIPTION
[0055] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.
[0056] The following describes a method for assessing the prognosis of lower limbs of stroke patients for brain-computer interface rehabilitation in an embodiment of the present application with reference to the accompanying drawings.
[0057] Figure 1 is a flow chart of a method for evaluating the prognosis of lower limbs of stroke patients for brain-computer interface rehabilitation according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:
[0058] Step 101 : extracting multimodal clinical medical data of stroke patients for BCI rehabilitation, wherein the multimodal clinical medical data includes demographic characteristics, clinical scale characteristics, and biomechanical characteristics.
[0059] In an embodiment of the present application, based on the hospital's electronic medical record platform, the electronic medical records of subacute stroke patients who undergo BCI rehabilitation training after admission are collected, a patient information database is established, and the multimodal clinical medical data of the stroke patients are extracted from the patient information database. The multimodal clinical medical data include demographic characteristics, clinical scale characteristics, and biomechanical characteristics.
[0060] like Figure 2 As shown, the demographic characteristics include age, gender, stroke type and disease duration; the clinical scale characteristics include motor ability, balance function, walking ability, daily living ability and spasticity level; the biomechanical characteristics include muscle strength level and horizontal range of motion.
[0061] More specifically, the three types of characteristics, demographics, clinical scales, and biomechanics, include: gender, age, stroke type, disease course, Fugl-Meyer lower limb motor function rating scale, modified Ashworth scale, Berg balance scale, stroke patient posture assessment scale, standing balance classification, 6-minute walk test, functional walking scale, modified Basel scale, dorsiflexion strength, toe flexor strength, dorsiflexion active range of motion, toe flexion active range of motion, dorsiflexion passive range of motion, and toe flexion passive range of motion.
[0062] It should be noted that demographic characteristics, clinical scale characteristics, and biomechanical-related characteristics were subsequently used to construct input features of the prediction model.
[0063] Step 102: Calculate the recovery rate of stroke patients according to the multimodal clinical medical data, and assign labels to stroke patients with different recovery rates.
[0064] In the embodiment of the present application, the recovery ratio of lower limb motor function of stroke patients is calculated based on the FMA-LE score before and after BCI rehabilitation training, and the expression is:
[0065] According to the Fugl-Meyer lower limb motor function score FMA-LE before and after BCI rehabilitation training, the recovery rate of lower limb motor function in stroke patients was calculated, and the expression was:
[0066]
[0067] Among them, FMA-LE pre 、FMA-LE post They refer to the FMA-LE scores before and after BCI rehabilitation training, respectively, maximumscore refers to the total score of the FMA-LE scale, and Recovery Proportion refers to the recovery proportion.
[0068] As a possible implementation, the total score of the FMA-LE scale is 34 points.
[0069] Then, by calculating the median recovery rate of all stroke patient samples included, the stroke patients with a recovery rate higher than the median were classified as responders, i.e., the better recovery group, and the stroke patients with a recovery rate lower than the median were classified as non-responders, i.e., the poor recovery group.
[0070] It should be noted that the partition labels and the recovery ratios are subsequently used to construct prediction features of the prediction model.
[0071] Step 103, standardizing the multimodal clinical medical data, and inputting the standardized multimodal clinical medical data into a trained elastic net model and an artificial neural network model ANN, respectively, to obtain a first prognosis evaluation result and a second prognosis evaluation result.
[0072] It is understandable that features need to be standardized before model training and prediction.
[0073] As a possible implementation method, the multimodal clinical medical data is standardized using a z-score standardization method so that its mean is zero and its variance is one, and the expression is:
[0074]
[0075] Among them, x ′ refers to the standardized features, x refers to the original features, mean refers to the average value of the original features of all included samples, and sd refers to the standard deviation of the original features of all included samples.
[0076] The following is a detailed description of the training process of the elastic network model Elastic net and the artificial neural network model ANN.
[0077] (1) The training process of the elastic network model Elastic net includes:
[0078] A leave-one-out cross-validation strategy was adopted to generate several different training sets and test sets from multimodal clinical medical data samples. In each split, in order to optimize the two hyperparameters (regularization parameter λ and mixing parameter ρ) of the Elastic net model in the training set, a second round of leave-one-out cross-validation was performed using the grid search method to search for the optimal hyperparameter set that produced the highest average classification accuracy in the leave-one-out cross-validation. Finally, the optimal hyperparameters were set on the entire training set to train the Elastic net model and predict the category of the test set.
[0079] In a possible embodiment, in the specific grid search optimization parameters, the regularization parameter λ value set is 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0, 2.1, 2.2, 2.3, 2.4, 2.5, 2.6, 2.7, 2.8, 2.9, 3.0, and the mixing parameter ρ value set is 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0.
[0080] (2) The training process of the artificial neural network model ANN includes:
[0081] A leave-one-out cross-validation strategy was adopted to generate several different training sets and test sets from multimodal clinical medical data samples. In each split, in order to optimize the number of hidden layer nodes m and the L2 regularization parameter α of the ANN model in the training set, a second round of leave-one-out cross-validation was performed using the grid search method to search for the optimal hyperparameter set that produced the highest average classification accuracy in the leave-one-out cross-validation. Finally, the optimal hyperparameters were set on the entire training set to train the ANN model and predict the categories of the test set.
[0082] In a possible embodiment, in the specific grid search optimization parameters, the value set of the number of hidden layer nodes m is 5, 6, 7, 8, 9, 10, and the value set of the L2 regularization parameter α is 0.0001, 0.001, 0.01, 0.1, 0.5.
[0083] In a possible embodiment, the maximum number of iterations is set to 1000, and the L-BFGS algorithm is used as the solver for weight optimization.
[0084] It should be noted that the multimodal clinical medical data samples are homologous to the multimodal clinical medical data after standardization, with only the individual feature values being different and the types being completely consistent.
[0085] Moreover, it can be understood that model training is an iterative process, which is carried out by continuously adjusting the network parameters of the model until the overall loss function value of the model is less than the preset value, or the overall loss function value of the model no longer changes or changes slowly, the model converges, and a trained model is obtained.
[0086] Optionally, the training may be considered finished when a preset number of training times is reached.
[0087] Optionally, the training may be considered finished when a preset training time is reached.
[0088] In addition, the prediction process using the trained elastic net model and artificial neural network model ANN is described in detail below.
[0089] (1) The process of inputting the standardized multimodal clinical medical data into the trained Elastic net for prediction includes: inputting the standardized multimodal clinical medical data into the trained Elastic net model for binary classification prediction of recovery ratio; applying L1 and L2 regularization based on the Elastic net model to predict the first FMA-LE score after BCI rehabilitation training through linear regression; calculating the first predicted recovery ratio of the lower limb motor function of the stroke patient according to the first FMA-LE score, and dividing the stroke patients with different first predicted recovery ratios according to the first predicted recovery ratio to obtain a first prognostic evaluation result.
[0090] It can be understood that the first prognostic assessment result classifies the patients into responders / non-responders for BCI rehabilitation, and the process of calculating the first FMA-LE score and calculating the first predicted recovery ratio of the lower limb motor function of the stroke patient based on the first FMA-LE score, and labeling the stroke patients with different first predicted recovery ratios based on the first predicted recovery ratio is analogous to step 102, and the present application will not repeat it in detail here.
[0091] (2) The process of inputting the standardized multimodal clinical medical data into the trained ANN for prediction includes:
[0092] The standardized multimodal clinical medical data is input into the trained ANN model for binary classification prediction of recovery ratio; based on the ANN model, the features in the input multimodal clinical medical data are aggregated and extracted, and the second FMA-LE score after BCI rehabilitation training is predicted through the sigmoid activation function; according to the second FMA-LE score, the second predicted recovery ratio of the lower limb motor function of the stroke patients is calculated, and according to the second predicted recovery ratio, the stroke patients with different second predicted recovery ratios are labeled to obtain a second prognostic evaluation result.
[0093] It should be noted that the ANN model is a multi-layer fully connected neural network, including an input layer, a hidden layer and an output layer.
[0094] It can be understood that the second prognostic assessment result classifies the patients into responders / non-responders for BCI rehabilitation, and the process of calculating the second FMA-LE score and calculating the second predicted recovery ratio of the lower limb motor function of the stroke patient based on the second FMA-LE score, and labeling the stroke patients with different second predicted recovery ratios based on the second predicted recovery ratio is analogous to step 102, and the present application will not repeat it in detail here.
[0095] It should be noted that the process of inputting the standardized multimodal clinical medical data into the trained Elasticnet and ANN for prediction is essentially the process of inputting demographic characteristics, clinical scale characteristics and biomechanical related characteristics into the trained Elastic net and ANN for prediction.
[0096] Step 104 , based on the actual result after the label division, the prediction capabilities of the Elastic net and the ANN are compared according to the first prognosis evaluation result and the second prognosis evaluation result to determine the optimal model.
[0097] In an embodiment of the present application, the true result after the label is divided and the first prognosis evaluation result are evaluated according to the evaluation index to obtain a first evaluation result, and the true result after the label is divided and the second prognosis evaluation result are evaluated according to the evaluation index to obtain a second evaluation result.
[0098] Among them, the evaluation indicators include but are not limited to classification accuracy, area under the curve, specificity and sensitivity.
[0099] Then, the first evaluation result is compared with the second evaluation result. If the first evaluation result is better than the second evaluation result, the Elastic net model is determined to be the optimal model. Otherwise, the ANN is determined to be the optimal model.
[0100] In a possible embodiment, Figure 3As shown in the figure, this is the confusion matrix result diagram of the Elastic net model and the ANN model. As can be seen from the figure, the classification accuracy of the Elastic net model is 0.810, the specificity is 0.778, and the sensitivity is 0.833; the classification accuracy of the ANN model is 0.762, the specificity is 0.722, and the sensitivity is 0.792.
[0101] Further, such as Figure 4 As shown in the figure, this is the ROC curve result diagram of the Elastic net model and the ANN model. It can be seen from the figure that the area under the curve of the Elastic net model is 0.782, and the area under the curve of the ANN model is 0.771.
[0102] Comprehensive comparison results show that the Elastic net model has better performance and can be determined as the optimal prediction model for lower limb prognosis of stroke patients for brain-computer interface rehabilitation.
[0103] Step 105: Based on the optimal model, a prognosis assessment is performed on the stroke patient for BCI rehabilitation.
[0104] This step is the application process of the optimal model.
[0105] by Figure 3 and Figure 4 Taking the illustrated embodiment as an example, the Elastic net model with better model performance is used as the optimal model to conduct prognosis assessment on stroke patients for BCI rehabilitation.
[0106] The embodiment of the present application extracts multimodal clinical medical data of stroke patients undergoing BCI rehabilitation and inputs the data into different models for evaluation to obtain the optimal model. The responsiveness to BCI rehabilitation therapy can be determined based on the patient's demographics, clinical scales, and biomechanical characteristics, thereby assisting physicians in customizing personalized therapies for patients.
[0107] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution of this application can be achieved, and this document is not limited here.
[0108] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.
Claims
1. A method for assessing the prognosis of lower limbs in stroke patients for brain-computer interface rehabilitation, comprising: Extracting multimodal clinical medical data of stroke patients for BCI rehabilitation, wherein the multimodal clinical medical data includes demographic characteristics, clinical scale characteristics, and biomechanical characteristics; Calculating the recovery rate of stroke patients according to the multimodal clinical medical data, and assigning labels to stroke patients with different recovery rates; Standardizing the multimodal clinical medical data, and inputting the standardized multimodal clinical medical data into a trained elastic network model Elasticnet and an artificial neural network model ANN, respectively, to obtain a first prognostic evaluation result and a second prognostic evaluation result; Based on the actual results after the label division, the prediction capabilities of Elasticnet and ANN are compared according to the first prognostic evaluation results and the second prognostic evaluation results to determine the optimal model; Based on the optimal model, the prognosis of stroke patients undergoing BCI rehabilitation is evaluated; Calculating the recovery ratio of the stroke patient according to the multimodal clinical medical data and assigning labels to stroke patients with different recovery ratios includes: According to the Fugl-Meyer lower limb motor function score FMA-LE before and after BCI rehabilitation training, the recovery rate of lower limb motor function in stroke patients was calculated, and the expression was: in, , Refers to the FMA-LE scores before and after BCI rehabilitation training, Refers to the total score of the FMA-LE scale, represents the recovery ratio; The median recovery rate of all stroke patient samples included was calculated, and the stroke patients with a recovery rate higher than the median were classified as responders, and the stroke patients with a recovery rate lower than the median were classified as non-responders; The training process of the elastic network model Elasticnet includes: The leave-one-out cross-validation strategy is adopted to generate several different training sets and test sets from multimodal clinical medical data samples; In each split, a second round of leave-one-out cross-validation is performed using a grid search method to search for the optimal set of hyperparameters that produces the highest average classification accuracy in leave-one-out cross-validation to optimize the regularization parameters of the Elasticnet model in the training set. and mixing parameters , Setting the optimal hyperparameters on the entire training set to train the Elastic net model and predict the category of the test set; The training process of the artificial neural network model ANN includes: The leave-one-out cross-validation strategy is adopted to generate several different training sets and test sets from multimodal clinical medical data samples; A second round of leave-one-out cross-validation was performed using a grid search method in each split to search for the optimal hyperparameter set that produced the highest average classification accuracy in the leave-one-out cross-validation to optimize the number of hidden layer nodes of the ANN model in the training set. and L2 regularization parameter ; Set the optimal hyperparameters on the entire training set to train the ANN model and predict the category of the test set; The method of comparing the prediction capabilities of Elasticnet and ANN based on the actual results after the label division and determining the optimal model according to the first prognostic evaluation result and the second prognostic evaluation result includes: Evaluating the true result after label division and the first prognosis evaluation result according to evaluation indicators to obtain a first evaluation result, wherein the evaluation indicators include classification accuracy, area under the curve, specificity and sensitivity; Evaluate the true result after labeling and the second prognosis evaluation result according to the evaluation index to obtain a second evaluation result; If the first evaluation result is better than the second evaluation result, the Elasticnet model is determined to be the optimal model; otherwise, the ANN is determined to be the optimal model.
2. The method according to claim 1, characterized in that: The multimodal clinical medical data of the stroke patient is extracted, including: Based on the hospital's electronic medical record platform, the electronic medical records of subacute stroke patients who received BCI rehabilitation training after admission were collected to establish a patient information database; The multimodal clinical medical data of stroke patients is extracted from the patient information database.
3. The method according to claim 1, characterized in that The demographic characteristics described included age, sex, stroke type, and disease duration; The clinical scale characteristics include motor ability, balance function, walking ability, daily living ability, and spasticity level; The biomechanical characteristics include muscle strength level and horizontal range of motion.
4. The method according to claim 1, characterized in that: The step of performing standardization processing on the multimodal clinical medical data comprises: The multimodal clinical medical data is standardized using the z-score standardization method to make its mean zero and variance one, and the expression is: in, Refers to the standardized characteristics, Refers to the original characteristics, Refers to the average value of all original features included in the sample, Refers to the standard deviation of the original characteristics of all included samples.
5. The method according to claim 3, characterized in that: The process of inputting the standardized multimodal clinical medical data into the trained Elasticnet for prediction includes: Inputting the standardized multimodal clinical medical data into the trained Elasticnet model to perform binary classification prediction of recovery ratio; Based on the Elastic net model, L1 and L2 regularization were applied simultaneously to predict the first FMA-LE score after BCI rehabilitation training through linear regression; According to the first FMA-LE score, the first predicted recovery ratio of the lower limb motor function of the stroke patient is calculated, and the stroke patients with different first predicted recovery ratios are labeled according to the first predicted recovery ratio to obtain a first prognostic assessment result.
6. The method according to claim 3, characterized in that The process of inputting the standardized multimodal clinical medical data into the trained ANN for prediction includes: Inputting the standardized multimodal clinical medical data into the trained ANN model to perform binary classification prediction of recovery ratio; Based on the ANN model, the features in the input multimodal clinical medical data are aggregated and extracted, and the second FMA-LE score after BCI rehabilitation training is predicted through a sigmoid activation function; According to the second FMA-LE score, the second predicted recovery ratio of the lower limb motor function of the stroke patients is calculated, and the stroke patients with different second predicted recovery ratios are labeled according to the second predicted recovery ratio to obtain a second prognostic assessment result.
