Tracking detection method and device for rehabilitation of apoplexy patient

Abstract

The invention particularly relates to a tracking detection method and device for rehabilitation of a stroke patient, and the method comprises the steps: collecting a brain blood oxygen signal through a few-lead near infrared spectrum device; after the signals are processed, near-infrared signals of key frequencies are extracted so as to construct a feature matrix and a weighted brain network topological graph; corresponding blood oxygen concentration data are merged into a three-dimensional matrix to be input into a bidirectional long-short-term memory network, a weighted brain network topological graph is input into a multi-core graph convolutional network, time sequence features and topological space features are extracted, a self-attention mechanism is introduced to dynamically adjust feature weights, and a patient rehabilitation state and trend evaluation model used for predicting the rehabilitation trend is constructed. Therefore, the problems that in the related technology, multi-lead near infrared spectrum equipment needs more sensors and more complex settings, a patient is required to be tested in a fixed environment, rehabilitation evaluation can not be provided for the patient anytime and anywhere, application of the multi-lead near infrared spectrum equipment in home rehabilitation is limited, and the cost of the multi-lead near infrared spectrum equipment is reduced are solved. Limitation exists in long-term rehabilitation monitoring of stroke patients.

Description

Technical Field

[0001] This application relates to the technical field of brain function detection and rehabilitation evaluation, and particularly relates to a tracking detection method and device for stroke patient rehabilitation. Background Art

[0002] Stroke is one of the main causes of death and disability globally, especially in the elderly population, bringing a heavy burden to patients and their families. Some studies have shown that local brain damage caused by stroke not only affects the function of the damaged area, but also disrupts the functional connectivity and network topology of other brain regions, affecting overall brain function. For stroke survivors, about 75% of patients have sequelae of varying degrees, and there is a window period of enhanced neural plasticity in the early stage of stroke rehabilitation. Improving the treatment effect in the early stage of rehabilitation can minimize the disability rate of patients. Therefore, the rehabilitation process of stroke patients usually requires long-term functional evaluation and tracking to timely adjust the treatment plan and improve the rehabilitation effect. fNIRS (functional near-infrared spectroscopy) is a non-invasive brain function imaging technology with characteristics such as portability, tolerance, non-invasiveness, cost-effectiveness, and long-term real-time monitoring ability, and has been widely used in neuroscience research.

[0003] In related technologies, stroke rehabilitation evaluation mostly relies on traditional devices and clinical evaluation scales in hospital or clinical environments. Traditional devices such as fMRI (functional magnetic resonance imaging) can provide relatively accurate brain function activity information, and fNIRS can measure the changes in oxygenated and deoxygenated hemoglobin in a similar way to fMRI for stroke rehabilitation evaluation.

[0004] However, in related technologies, multi-channel fNIRS devices require more sensors and more complex settings, require patients to be tested in a fixed environment, cannot provide rehabilitation evaluation for patients anytime and anywhere, limit its application in home rehabilitation, and have certain limitations in the long-term rehabilitation monitoring of stroke patients, which need to be solved urgently. Summary of the Invention

[0005] This application provides a tracking detection method and device for stroke patient rehabilitation to solve the problems in related technologies that multi-channel fNIRS devices require more sensors and more complex settings, require patients to be tested in a fixed environment, cannot provide rehabilitation evaluation for patients anytime and anywhere, limit its application in home rehabilitation, and have certain limitations in the long-term rehabilitation monitoring of stroke patients.

[0006] The first aspect of the present application provides a tracking and detection method for stroke patient rehabilitation, including the following steps: obtaining the cerebral blood oxygen signal of a stroke patient in the rehabilitation period collected by an oligolead near-infrared spectroscopy device; preprocessing and amplifying the cerebral blood oxygen signal to obtain processed data, and extracting near-infrared signals of multiple target key frequencies from the processed data to construct a feature matrix, and constructing a weighted brain network topology map based on at least one of the feature matrices; merging the blood oxygen concentration data corresponding to the cerebral blood oxygen signal into a three-dimensional matrix to input into a bidirectional long short-term memory network, and inputting the weighted brain network topology map into a multi-core graph convolutional network to extract temporal features and topological spatial features respectively, and mapping the temporal features, the topological spatial features and manual features to a unified dimension and then splicing and fusing them, introducing a self-attention mechanism to dynamically adjust the feature weights, and constructing a patient rehabilitation state and trend evaluation model for predicting the rehabilitation trend.

[0007] Optionally, in an embodiment of the present application, the preprocessing and amplifying the cerebral blood oxygen signal to obtain processed data, and extracting near-infrared signals of multiple target key frequencies from the processed data to construct a feature matrix, and constructing a weighted brain network topology map based on the feature matrix includes: converting the cerebral blood oxygen signal into the blood oxygen concentration data, and removing systematic chronic drift, removing jump noise, and removing irrelevant physiological noise and low-frequency noise to obtain initial processed data; decomposing and reconstructing the initial processed data based on wavelet transform to extract different frequency band sequences; enhancing the data by means of data segmentation and reconstruction based on the different frequency band sequences to obtain the processed blood oxygen signal; calculating various time-frequency features and frequency-specific functional connection features based on wavelet phase coherence for the processed blood oxygen signal to generate the feature matrix, and constructing a weighted brain network topology map based on at least one of the feature matrices.

[0008] Optionally, in an embodiment of the present application, the output layer of the patient rehabilitation state evaluation model includes a regression task branch and a classification task branch. Among them, the regression task branch is used to quantitatively evaluate the rehabilitation state of the patient, the classification task branch is used to predict the rehabilitation trend of the patient, the regression task branch uses the mean square error as the loss function, and the classification task branch uses the cross entropy as the loss function.

[0009] Optionally, in an embodiment of the present application, the constructing a patient rehabilitation state and trend evaluation model for predicting the rehabilitation trend includes: optimizing the joint loss function through a multi-task optimization strategy, and dynamically adjusting the weight factors of the regression task and the classification task to balance the optimization priorities of the tasks.

[0010] Optionally, in an embodiment of the present application, where

[0011] The optimization formula of the combined loss function is as follows:

[0012]

[0013] where L regression and L classification are the loss functions of the regression task and the classification task respectively, and σ reg and σ cia are the weight factors of the regression task and the classification task respectively;

[0014] The adjustment formula of the weight factor is as follows:

[0015]

[0016]

[0017] where L regression and L classification are the loss functions of the regression task and the classification task respectively, ε0 is the initial smoothing term to avoid numerical explosion caused by the denominator being zero, and γ is the decay coefficient.

[0018] In the second aspect of the embodiments of the present application, a tracking and detection device for stroke patient rehabilitation is provided, including: an acquisition module, configured to acquire the cerebral blood oxygen signal of a stroke patient in the rehabilitation period collected by an oligolead near-infrared spectroscopy device; a processing module, configured to preprocess and amplify the cerebral blood oxygen signal to obtain processed data, extract near-infrared signals of multiple target key frequencies from the processed data to construct a feature matrix, and construct a weighted brain network topology map based on at least one of the feature matrices; a prediction module, configured to merge the blood oxygen concentration data corresponding to the cerebral blood oxygen signal into a three-dimensional matrix to input a bidirectional long short-term memory network, and input the weighted brain network topology map into a multi-core graph convolutional network, extract temporal features and topological space features respectively, and splice and fuse the temporal features, the topological space features and manual features after mapping them to a unified dimension, introduce a self-attention mechanism to dynamically adjust the feature weights, and construct a patient rehabilitation status and trend evaluation model for predicting the rehabilitation trend.

[0019] Optionally, in an embodiment of the present application, the processing module includes: a conversion unit configured to convert the cerebral blood oxygen signal into the blood oxygen concentration data, and remove systematic chronic drift, remove jump noise, and remove irrelevant physiological noise and low-frequency noise to obtain initial processed data; an extraction unit configured to decompose and reconstruct the initial processed data based on wavelet transform to extract different frequency band sequences; an enhancement unit configured to enhance the data in a data segmentation and reconstruction manner based on the different frequency band sequences to obtain a processed blood oxygen signal; a generation unit configured to calculate various time-frequency features and frequency-specific functional connection features based on wavelet phase coherence for the processed blood oxygen signal, generate the feature matrix, and construct a weighted brain network topology map based on at least one of the feature matrix.

[0020] Optionally, in an embodiment of the present application, the output layer of the patient rehabilitation status evaluation model includes a regression task branch and a classification task branch. Among them, the regression task branch is used to quantitatively evaluate the patient's rehabilitation status, the classification task branch is used to predict the patient's rehabilitation trend, the regression task branch uses the mean square error as the loss function, and the classification task branch uses the cross entropy as the loss function.

[0021] Optionally, in an embodiment of the present application, the prediction module includes: an adjustment unit configured to optimize the joint loss function through a multi-task optimization strategy and dynamically adjust the weight factors of the regression task and the classification task to balance the optimization priorities of the tasks.

[0022] Optionally, in an embodiment of the present application, where

[0023] The optimization formula of the joint loss function is:

[0024]

[0025] Where L regression and L classification are the loss functions of the regression task and the classification task respectively, and σ reg and σ cia are the weight factors of the regression task and the classification task respectively;

[0026] The adjustment formula of the weight factor is:

[0027]

[0028]

[0029] Where L regression and L classificationare the loss functions for the regression task and the classification task respectively, ε0 is the initial smoothing term to avoid numerical explosion caused by a zero denominator, and γ is the decay coefficient.

[0030] An embodiment of the third aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the program to implement the tracking detection method for stroke patient rehabilitation as described in the above embodiment.

[0031] An embodiment of the fourth aspect of the present application provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the program is executed by a processor, it implements the tracking detection method for stroke patient rehabilitation as described above.

[0032] An embodiment of the fifth aspect of the present application provides a computer program product, where the computer program product stores a computer program, and when the program is executed by a processor, it implements the tracking detection method for stroke patient rehabilitation as described above.

[0033] Embodiments of the present application can collect the cerebral blood oxygen signals of stroke patients through a portable oligomeric-lead fNIRS device to construct a feature matrix, and based on at least one of the feature matrices, construct a weighted brain network topology map to input into a deep learning model. Through a multi-task learning model and an adaptive feature fusion mechanism, temporal features, topological features, and manual features are dynamically integrated, and then a rehabilitation score and a trend prediction result can be output, thereby establishing a precise tracking detection method for stroke patient rehabilitation, and further realizing early intervention and personalized treatment during the rehabilitation process of stroke patients. Thus, it solves the problems in the related art that multi-lead fNIRS devices require more sensors and more complex settings, require patients to be tested in a fixed environment, cannot provide rehabilitation evaluations for patients anytime and anywhere, limit their application in home rehabilitation, and have certain limitations in the long-term rehabilitation monitoring of stroke patients.

[0034] Additional aspects and advantages of the present application will be given in part in the following description, become apparent in part from the following description, or be understood through the practice of the present application. Description of the Drawings

[0035] The above and / or additional aspects and advantages of the present application will become apparent and be easily understood from the following description of the embodiments in conjunction with the drawings, where:

[0036] Figure 1 is a flowchart of a tracking detection method for stroke patient rehabilitation according to an embodiment of the present application;

[0037] Figure 2Schematic diagram of merging oligochannel HbO (Oxyhemoglobin) and HbR (Deoxyhemoglobin) data into a three-dimensional matrix provided according to an embodiment of the present application;

[0038] Figure 3 Schematic diagram of the FM (Fugl-Meyer Assessment) score and rehabilitation trend assessment method for rehabilitation monitoring provided according to an embodiment of the present application;

[0039] Figure 4 Schematic diagram of the structure of a tracking detection device for stroke patient rehabilitation provided according to an embodiment of the present application;

[0040] Figure 5 Schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. Detailed implementation manners

[0041] The embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application and should not be construed as a limitation to the present application.

[0042] The tracking detection method and device for stroke patient rehabilitation according to the embodiments of the present application will be described below with reference to the accompanying drawings. In view of the problems in the related technologies mentioned in the above background technology, that is, multi-channel fNIRS devices require more sensors and more complex settings, require patients to be tested in a fixed environment, cannot provide rehabilitation assessment for patients at any time and anywhere, limit their application in home rehabilitation, and have certain limitations in the long-term rehabilitation monitoring of stroke patients, the present application provides a tracking detection method for stroke patient rehabilitation. In this method, the cerebral blood oxygen signal of stroke patients can be collected by a portable oligochannel fNIRS device to construct a feature matrix, and a weighted brain network topology map can be constructed based on at least one of the feature matrices and input into a deep learning model. Through a multi-task learning model and an adaptive feature fusion mechanism, temporal features, topological features and manual features can be dynamically integrated, and then a rehabilitation score and a trend prediction result can be output, so as to establish a precise tracking detection rehabilitation assessment and prediction method for stroke patient rehabilitation, and further realize early intervention and personalized treatment in the rehabilitation process of stroke patients. Thus, the problems in the related technologies, such as multi-channel fNIRS devices requiring more sensors and more complex settings, requiring patients to be tested in a fixed environment, unable to provide rehabilitation assessment for patients at any time and anywhere, limiting their application in home rehabilitation, and having certain limitations in the long-term rehabilitation monitoring of stroke patients, are solved.

[0043] Specifically, Figure 1 is a schematic flowchart of a tracking detection method for stroke patient rehabilitation provided by an embodiment of the present application.

[0044] As Figure 1 shown, the tracking detection method for stroke patient rehabilitation includes the following steps:

[0045] In step S101, acquire the cerebral oxygenation signal of a stroke patient in the rehabilitation period collected by an oligolead near-infrared spectroscopy device.

[0046] It can be understood that in the embodiment of the present application, signals in the frontal region can be mainly collected, covering the relevant brain regions for cognition and motor control.

[0047] For example, in the embodiment of the present application, a portable oligolead fNIRS device can be used to monitor the brain function of a stroke patient in the rehabilitation period. The light source wavelengths are 735nm and 850nm, the sampling rate is 10Hz, and 8 sensors, that is, 8 leads, are set in the frontal region of the patient. The layout covers the patient's DLPFC (Dorsolateral Prefrontal Cortex) and PMC (Premotor cortex). By irradiating the patient's head with near-infrared light and detecting the reflected light, the cerebral oxygenation signal in the frontal region of the patient is mainly collected, and the concentration changes of HbO and deoxyhemoglobin HbR are recorded.

[0048] In the embodiment of the present application, through a portable oligolead fNIRS device, signals in the frontal region of the patient can be mainly collected, covering the relevant brain regions for cognition and motor control, ensuring the efficiency and portability of data collection.

[0049] In step S102, preprocess and amplify the cerebral oxygenation signal to obtain processed data, extract near-infrared signals of multiple target key frequencies from the processed data to construct a feature matrix, and construct a weighted brain network topology map based on at least one of the feature matrices.

[0050] It can be understood that in the embodiment of the present application, the near-infrared signals of multiple target key frequencies can be respectively frequency band Ⅰ (0.145Hz - 0.6Hz), Ⅱ (0.052Hz - 0.145Hz), Ⅲ (0.021Hz - 0.052Hz), Ⅳ (0.0095Hz - 0.021Hz). In physiological research, the physiological meanings of the cerebral oxygenation signals in these frequency bands are respectively respiratory activity, myogenic activity, neurogenic activity, and endothelial cell metabolic activity related to nitric oxide response.

[0051] In the actual execution process, the embodiments of the present application can eliminate interference factors in the signal through a preprocessing method, so as to obtain relatively stable and accurate brain function processing data, decompose and reconstruct the fNIRS signal, extract near-infrared signals in multiple frequency intervals from the processing data, construct a weighted functional connectivity matrix, and further construct a weighted brain network topology map.

[0052] The embodiments of the present application can achieve a comprehensive and accurate modeling of brain function signals through the preprocessing, feature extraction, and construction of a weighted brain network topology map of cerebral blood oxygen signals, and enhance the prediction ability of the model.

[0053] Optionally, in an embodiment of the present application, the cerebral blood oxygen signal is preprocessed and amplified to obtain processing data, and near-infrared signals of multiple target key frequencies are extracted from the processing data to construct a feature matrix, and a weighted brain network topology map is constructed based on at least one of the feature matrices, including: converting the cerebral blood oxygen signal into blood oxygen concentration data, and removing systematic chronic drift, removing jump noise, and removing irrelevant physiological noise and low-frequency noise to obtain initial processing data; decomposing and reconstructing the initial processing data based on wavelet transform to extract different frequency band sequences; enhancing the data by means of data segmentation and reconstruction based on different frequency band sequences to obtain the processed blood oxygen signal; calculating various time-frequency features and frequency-specific functional connectivity features based on wavelet phase coherence for the processed blood oxygen signal to generate a feature matrix, and constructing a weighted brain network topology map based on at least one of the feature matrices.

[0054] It can be understood that, in the embodiments of the present application, the irrelevant physiological noise and low-frequency noise may be noise generated by physiological activities such as heartbeat (about 1 Hz) in the fNIRS signal and low-frequency physiological signal (less than 0.01 Hz) noise generated by some endothelial cell metabolic activities not related to nitric oxide reaction.

[0055] For example, the embodiments of the present application may adopt the modified Lambert-Beer law to convert the collected original optical signal data into blood oxygen concentration data of HbO, HbR, and HbT (Total Hemoglobin), set the DPF (Differential Pathlength Factor) to 5.3 (735 nm) and 4.9 (850 nm), use a fifth-order polynomial fitting to eliminate the system baseline drift, perform head motion correction using TDDR (Temporal Derivative Distribution Repair), repair mutation noise, calculate the signal derivative distribution within a sliding window with a width of 10 seconds, and if the derivative at a certain point exceeds the mean ± 3 standard deviations, repair it with linear interpolation of the five points before and after. A band-pass filter with a frequency range of 0.01 - 0.2 Hz is used to remove heartbeat noise and low-frequency drift, remove systematic chronic drift, remove jump noise, and remove irrelevant physiological noise and low-frequency noise to obtain the initial processed data.

[0056] Further, a multi-band wavelet decomposition method is used to extract frequency band sequences related to different physiological activities, such as but not limited to: 0.145 Hz - 0.6 Hz, 0.052 Hz - 0.145 Hz, 0.021 Hz - 0.052 Hz, 0.0095 Hz - 0.021 Hz.

[0057] Specifically, next, the preprocessed signal is subjected to data augmentation using the method of segmentation and reconstruction. For example, the window length can be set to 30 seconds and the overlap rate to 50%. The signals of different motion tasks of the same patient are spliced to enhance the utilization rate of temporal information and the diversity of training samples. After data augmentation, the obtained HbO and HbR signals are synthesized into a three-dimensional matrix. At each sampling time point, a 2*8 two-dimensional matrix will be formed, representing 2 types of signals * 8 leads. The training set, validation set, and test set are divided, and k-fold cross-validation is used to avoid data distribution deviation. Based on different frequency band sequences, data is enhanced by means of data segmentation and reconstruction to obtain the processed blood oxygen signal.

[0058] Furthermore, various time-frequency features and frequency-specific functional connectivity features based on wavelet phase coherence are calculated for the processed blood oxygen signals to generate a feature matrix. For example, a manual feature extraction method is combined with deep learning to extract some traditional statistical features from HbO and HbR signals respectively, including mean, variance, peak value, integral, root mean square, etc., as well as features such as power spectral density, primary and secondary fitting coefficients, approximate entropy, and frequency band energy ratio; nonlinear dynamic features such as Lyapunov exponent, sample entropy, and fractal dimension are calculated in the hybrid feature framework to reflect the complexity and irregularity of the signals, and multi-dimensional correlation features are calculated to measure the coupling properties and functional connectivity relationships between different channels of the signals through Pearson correlation coefficient. For example, the dynamic coupling of HbO and HbR signals is analyzed through Pearson correlation coefficient, and the cross-channel volatility difference is extracted. These manual features help to capture the basic patterns of the signals; the manually extracted features are screened and normalized. In addition, a weighted brain network topology map is constructed based on at least one of the feature matrices. For example, the WPCO (Weighted Phase Consistency) between 8 leads is calculated, and the brain network topology map is constructed through frequency-specific functional connectivity. The nodes are set as fNIRS channels, and the edge weights are WPCO values.

[0059] In the embodiments of the present application, through the preprocessing, multi-band decomposition, data augmentation, time-frequency feature extraction and construction of the weighted brain network topology map of the cerebral blood oxygen signals, the accurate modeling of the brain function signals is realized, and the prediction ability of the model is enhanced through multi-dimensional feature integration and adaptive feature fusion mechanism.

[0060] In step S103, the blood oxygen concentration data corresponding to the cerebral blood oxygen signals are merged into a three-dimensional matrix to be input into a bidirectional long short-term memory network, and the weighted brain network topology map is input into a multi-kernel graph convolutional network to extract temporal features and topological spatial features respectively. The temporal features, topological spatial features and manual features are mapped to the same dimension and then spliced and fused, and a self-attention mechanism is introduced to dynamically adjust the feature weights to construct a patient rehabilitation state and trend evaluation model for predicting the rehabilitation trend.

[0061] It can be understood that the embodiments of the present application can construct an innovative multi-task learning framework BLSM-Net to establish an adaptive feature fusion network model bidirectional long short-term memory network based on Bi-LSTM (Bidirectional Long Short-Term Memory Network) and SE-MGCN (Squeeze-and-Excitation-Multi-Kernel Graph Convolutional Network).

[0062] In the actual execution process, in the deep learning feature extraction of the embodiments of the present application, the time series data and the brain network graph can be used as two inputs respectively, and processed through parallel network branches. The time series data captures the time series characteristics of the signal through the Bi-LSTM module, and the brain network graph extracts the topological information of the brain network through the SE-MGCN module; the manually input features enter the feature fusion layer, and these two parts of features are concatenated and fused after being mapped to the same dimension, and weights are adaptively assigned to the features from different sources to obtain an integrated feature representation, and the time series features, topological space features and manually input features are mapped to the same dimension and then concatenated and fused.

[0063] Further, in the deep learning modeling of the embodiments of the present application, a multi-task learning framework BLSM-Net can be adopted, and its architecture can but is not limited to include Bi-LSTM, SE-MGCN and an adaptive feature fusion module. The input layer can synthesize HbO and HbR signals into a blood oxygen matrix of size 2×8×3000 (8 channels, 3000 sampling points) and input it into the Bi-LSTM network, the weighted brain network topology graph can be input into the SE-MGCN network, and the manually input feature matrix can be directly input into the fusion layer; the Bi-LSTM module can capture the dynamic change trend and periodic pattern of the signal through the bidirectional LSTM (Long Short-Term Memory Network) layer, and multiple layers of LSTM are stacked to learn the long-term and short-term dependencies of the signal in the time dimension layer by layer, and the number of LSTM units can be used as an adjustable parameter. For example, the initial number of hidden units is set to 64; the SE-MGCN module uses multi-kernel graph convolution with kernel sizes of 3, 5, and 7 to extract local neighborhoods, medium-range connections and global topological features of the functional connections, and can dynamically enhance the weights of key channels through the squeeze-and-excitation mechanism; in the feature fusion stage, the time series features output by the Bi-LSTM, the topological features output by the SE-MGCN and the manually input features are mapped to the same dimension, and the attention mechanism is used to calculate the weights, and then enter the multi-task branches, including the regression task branch and the classification task branch, so as to construct a patient rehabilitation state and trend evaluation model for predicting the rehabilitation trend.

[0064] Furthermore, the embodiments of the present application can perform model evaluation. After training is completed, an independent test set is used to evaluate the model. The evaluation metrics can include, but are not limited to, using MSE (Mean Squared Error) and RMSE (Root Mean Squared Error) to evaluate the performance of regression tasks, and using accuracy, F1 score, ROC curve, and AUC (Area Under Curve) to evaluate the performance of classification tasks. These evaluation metrics can comprehensively measure the performance of the model in stroke patient rehabilitation prediction and further adjust the model parameters to achieve the optimal effect.

[0065] The embodiments of the present application can achieve accurate evaluation and trend prediction of the rehabilitation status of stroke patients by combining cerebral blood oxygen signals into a three-dimensional matrix and inputting it into Bi-LSTM, inputting the weighted brain network topology map into SE-MGCN, and combining an adaptive feature fusion mechanism. The trained deep learning model will be able to predict the patient's rehabilitation process based on the input fNIRS signals, providing valuable rehabilitation evaluation basis for doctors and helping patients achieve personalized rehabilitation management.

[0066] Optionally, in an embodiment of the present application, the output layer of the patient rehabilitation status evaluation model includes a regression task branch and a classification task branch. Among them, the regression task branch is used to quantitatively evaluate the patient's rehabilitation status, and the classification task branch is used to predict the patient's rehabilitation trend. The regression task branch uses mean squared error as the loss function, and the classification task branch uses cross-entropy as the loss function.

[0067] It can be understood that cross-entropy in the embodiments of the present application is a measure used in machine learning and statistics to measure the difference between two probability distributions, and it can be used as a loss function in classification problems.

[0068] In the actual execution process, the output layer in the embodiments of the present application includes a regression task branch and a classification task branch. The regression task branch can quantitatively evaluate the patient's rehabilitation status. Two fully connected layers are set up to first compress and extract key features, and then map the features back to the quantified rehabilitation degree score, using MSE as the loss function. Finally, the rehabilitation score is output based on Fugl-Meyer, and the output range is 0-100. The classification task branch can classify and predict the patient's future rehabilitation trend. Using a fully connected layer, it non-linearly projects the global features and uses the Softmax activation function to generate the class probability distribution, corresponding to three categories: improved rehabilitation, maintained rehabilitation, and deteriorated rehabilitation, and uses cross-entropy as the loss function to predict the rehabilitation trend, outputting the probability distribution results of the three categories: improved, maintained, and deteriorated.

[0069] The embodiments of the present application utilize the advantages of deep learning technology to provide more accurate and comprehensive rehabilitation status evaluation and trend prediction.

[0070] Optionally, in an embodiment of the present application, a patient rehabilitation status and trend assessment model for predicting the rehabilitation trend is constructed, including: optimizing the joint loss function through a multi-task optimization strategy, and dynamically adjusting the weight factors of the regression task and the classification task to balance the optimization priorities of the tasks.

[0071] It can be understood that the embodiments of the present application can adopt k-fold cross-validation (training set: validation set = 8:2) and a dynamic weight adjustment strategy. Hyperparameter tuning includes the convolution kernel size, the number of LSTM units, etc. By dynamically updating the weight factors in the joint loss function, the optimization priorities of the regression task and the classification task are balanced. During the optimization process, the Adam optimization algorithm is used to update the model parameters, overfitting and underfitting problems are concerned, and tuning is performed through means such as regularization, early stopping strategy, and cross-validation, and combined with a learning rate scheduling strategy to avoid the model falling into a local optimal solution, thereby improving the training efficiency and model performance.

[0072] The embodiments of the present application can improve the overall performance and adaptability of the model by a multi-task optimization strategy and dynamically adjusting the weight factors of the regression task and the classification task, and enhance its flexibility and practicality in practical applications.

[0073] Optionally, in an embodiment of the present application, wherein

[0074] The optimization formula of the joint loss function is:

[0075]

[0076] Wherein, L regression and L classification are the loss functions of the regression task and the classification task respectively, σ reg and σ cia are the weight factors of the regression task and the classification task respectively;

[0077] The adjustment formula of the weight factor is:

[0078]

[0079] Wherein, L regression and L classification are the loss functions of the regression task and the classification task respectively, ε0 is the initial smoothing term to avoid numerical explosion caused by the denominator being zero, and γ is the attenuation coefficient.

[0080] During the actual execution process, the embodiments of the present application can adopt a multi-task optimization strategy to effectively coordinate the two task branches, define a joint loss function to optimize the model:

[0081]

[0082] Among them, L regression and L cladsification are the loss functions for the regression task and the classification task respectively, and σ reg and σ cia are the weight factors for the regression task and the classification task respectively;

[0083] Furthermore, a mechanism based on dynamic loss weighting is introduced. By adjusting the weight factors of the regression task and the classification task, the optimization priorities of the two tasks are dynamically balanced. After each training iteration, the weight factors are adjusted according to the ratio of the loss values of the two tasks, and the joint loss function is updated to guide the optimization direction, preventing the imbalance of the dominant task and promoting the learning of the low-loss task:

[0084]

[0085]

[0086] Among them, L regression and L classification are the loss functions for the regression task and the classification task respectively, ε0 is the initial smoothing term to avoid numerical explosion caused by the denominator being zero, γ is the decay coefficient, which decreases gradually with the increase of the training round t, and the smoothing term depends on the dynamic adjustment of the task loss in the later stage.

[0087] The embodiments of the present application can adopt the strategies of the joint loss function optimization formula and the weight factor adjustment formula, improving the overall performance of the model and promoting the personalized rehabilitation management and treatment of stroke patients.

[0088] Specifically, as shown in combination with Figure 2 and Figure 3 , the principle of the tracking and detection method for stroke patient rehabilitation in the embodiments of the present application is elaborated in detail with a specific embodiment.

[0089] Figure 2 is the process of combining the HbO and HbR signals of oligochannel (8 leads) into a three-dimensional matrix.

[0090] The embodiments of the present application can utilize the three-dimensional matrix to simultaneously retain the temporal information, spatial information (different channels), and signal type information of the signal, providing rich input data for the model. This data representation method enables Bi-LSTM to effectively capture the temporal dependence relationship in the cerebral blood oxygen signal, extract the dynamic change trend of the signal, and by retaining the channel dimension, the model can capture the functional connection changes between different brain regions, providing a basis for subsequent topological feature extraction.

[0091] Figure 3It is a schematic diagram of the FM score and rehabilitation trend assessment method for rehabilitation monitoring. An innovative multi-task learning framework BLSM-Net is proposed in the embodiments of this application, and an adaptive feature fusion network model based on Bi-LSTM and SE-MGCN is established. The embodiments of this application may include the following steps:

[0092] Step S301: Input layer: Input the HbO-HbT 3D matrix constructed from the original fNIRS data, the weighted brain network topology map constructed from the functional connectivity, and the manually extracted mixed feature matrix into the deep learning model.

[0093] Among them, in the embodiments of this application, the preprocessed HbO signal and HbR signal can be normalized and then merged into a three-dimensional matrix of (2*C*T), where 2 represents two signal types, C is the number of channels, and T is the number of sampling points.

[0094] Step S302: Feature extraction layer: The blood oxygen matrix is input into the Bi-LSTM module to capture the long-term dependencies in both directions of the sequence; the brain network topology map is input into the SE-MGCN module, and the multi-core graph convolutional network with squeeze and excitation mechanisms is used to extract the brain topology structure information.

[0095] Among them, in the embodiments of this application, the blood oxygen matrix can be input into the Bi-LSTM module, the weighted brain network topology map is input into the SE-MGCN module, and the manual feature matrix is input into the feature fusion layer; the Bi-LSTM module and the SE-MGCN module form a parallel structure to jointly extract key features. During the stroke rehabilitation process, the brain function changes of patients often show a long-term and progressive recovery pattern. The Bi-LSTM module adopts a bidirectional LSTM mechanism, and through its gating mechanism, it extracts important long-term dependencies in the sequence data, providing deep temporal features for subsequent prediction; the SE-MGCN module extracts local and global topology features of the functional connectivity brain network through the multi-core graph convolutional network, and dynamically enhances the weights of important channels through the squeeze and excitation (SE) mechanism.

[0096] Step S303: Feature fusion layer: After mapping the manually extracted mixed feature matrix and the feature matrices extracted by the parallel Bi-LSTM and SE-MGCN to the same dimension and normalizing them, they are concatenated and fused, and an attention mechanism is introduced to dynamically adjust the feature weights.

[0097] Among them, in the embodiments of this application, the feature matrices extracted by Bi-LSTM and SE-MGCN and the manual feature matrix can be mapped to the same dimension, and the attention mechanism is used to calculate the weights, and after dynamic fusion, they are input into the fully connected layer.

[0098] Step S304: Adopt a multi-task optimization strategy, introduce a mechanism based on dynamic loss weighting, and dynamically balance the optimization priorities of the two tasks by adjusting the weight factors of the regression task and the classification task.

[0099] Specifically, the embodiments of the present application can introduce a mechanism based on dynamic loss weighting, dynamically balance the optimization priorities of the two tasks by adjusting the weight factors of the regression task and the classification task. After each training iteration, adjust the weight factors according to the loss value ratio of the two tasks, update the joint loss function to guide the optimization direction, prevent the dominant task from being unbalanced, and promote the learning of the low-loss task.

[0100] Step S305: Output layer: The regression task branch outputs the predicted FM score, and the regression task branch outputs the predicted rehabilitation trend.

[0101] Specifically, the embodiments of the present application can include multi-task branches, including a regression task branch, which quantitatively evaluates the rehabilitation status of patients, sets two fully connected layers, first compresses and extracts key features, and then maps the features back to the quantified rehabilitation degree score, and uses MSE as the loss function; a classification task branch, which classifies and predicts the future rehabilitation trend of patients, uses a fully connected layer to perform a non-linear projection on the global features, and uses the Softmax activation function to generate the class probability distribution, corresponding to three categories: improved rehabilitation, maintained rehabilitation, and deteriorated rehabilitation, and uses cross-entropy as the loss function.

[0102] According to the stroke patient rehabilitation tracking and detection method proposed by the embodiments of the present application, the cerebral oxygenation signals of stroke patients can be collected through a portable multi-channel fNIRS device, and a weighted brain network topology map can be constructed through a hybrid feature extraction framework and input into a deep learning model. Through a multi-task learning model and an adaptive feature fusion mechanism, temporal features, topological features, and manual features can be dynamically integrated, and then the rehabilitation score and trend prediction results can be output, so as to establish a precise stroke patient rehabilitation tracking and detection rehabilitation evaluation and prediction method, and further realize early intervention and personalized treatment during the rehabilitation process of stroke patients. Thus, it solves the problems in the related art that multi-channel fNIRS devices require more sensors and more complex settings, require patients to be tested in a fixed environment, cannot provide rehabilitation evaluation for patients at any time and anywhere, limit their application in home rehabilitation, and have certain limitations in the long-term rehabilitation monitoring of stroke patients.

[0103] Next, refer to the drawings to describe the stroke patient rehabilitation tracking and detection device according to the embodiments of the present application.

[0104] Figure 4 It is a block diagram of the stroke patient rehabilitation tracking and detection device according to the embodiments of the present application.

[0105] As Figure 4As shown, the tracking and detection device 10 for stroke patient rehabilitation includes: an acquisition module 100, a processing module 200, and a prediction module 300.

[0106] Among them, the acquisition module 100 is used to acquire the cerebral blood oxygen signal of a stroke patient in the rehabilitation period collected by an oligolead near-infrared spectroscopy device.

[0107] The processing module 200 is used to preprocess and amplify the cerebral blood oxygen signal to obtain processed data, extract near-infrared signals of multiple target key frequencies from the processed data to construct a feature matrix, and construct a weighted brain network topology map based on the feature matrix.

[0108] The prediction module 300 is used to merge the blood oxygen concentration data corresponding to the cerebral blood oxygen signal into a three-dimensional matrix to input into a bidirectional long short-term memory network, and input the weighted brain network topology map into a multi-core graph convolutional network to extract temporal features and topological spatial features respectively, and map the temporal features, topological spatial features, and manual features to the same dimension and then splice and fuse them, introduce a self-attention mechanism to dynamically adjust the feature weights, and construct a patient rehabilitation status and trend evaluation model for predicting the rehabilitation trend.

[0109] Optionally, in an embodiment of the present application, the processing module 200 includes: a conversion unit, an extraction unit, an enhancement unit, and a generation unit.

[0110] Among them, the conversion unit is used to convert the cerebral blood oxygen signal into blood oxygen concentration data, and remove systematic chronic drift, remove jump noise, and remove irrelevant physiological noise and low-frequency noise to obtain initial processed data.

[0111] The extraction unit is used to decompose and reconstruct the initial processed data based on wavelet transform to extract different frequency band sequences.

[0112] The enhancement unit is used to enhance the data based on different frequency band sequences by means of data segmentation and reconstruction to obtain the processed blood oxygen signal.

[0113] The generation unit is used to calculate various time-frequency features and frequency-specific functional connection features based on wavelet phase coherence for the processed blood oxygen signal, generate a feature matrix, and construct a weighted brain network topology map based on at least one of the feature matrix.

[0114] Optionally, in an embodiment of the present application, the output layer of the patient rehabilitation status evaluation model includes a regression task branch and a classification task branch. Among them, the regression task branch is used to quantitatively evaluate the patient's rehabilitation status, the classification task branch is used to predict the patient's rehabilitation trend, the regression task branch uses the mean square error as the loss function, and the classification task branch uses the cross entropy as the loss function.

[0115] Optionally, in an embodiment of the present application, the prediction module 300 includes: an adjustment unit.

[0116] The adjustment unit is configured to optimize the joint loss function through a multi-task optimization strategy and dynamically adjust the weight factors of the regression task and the classification task to balance the optimization priorities of the tasks.

[0117] Optionally, in an embodiment of the present application, where

[0118] The optimization formula of the joint loss function is:

[0119]

[0120] where L regression and L classification are the loss functions of the regression task and the classification task respectively, and σ reg and σ cia are the weight factors of the regression task and the classification task respectively;

[0121] The adjustment formula of the weight factor is:

[0122]

[0123]

[0124] where L regression and L classification are the loss functions of the regression task and the classification task respectively, ε0 is the initial smoothing term to avoid numerical explosion caused by the denominator being zero, and γ is the attenuation coefficient.

[0125] It should be noted that the foregoing explanation of the embodiment of the tracking detection method for stroke patient rehabilitation also applies to the tracking detection device for stroke patient rehabilitation in this embodiment, and will not be elaborated here.

[0126] The tracking and detection device for stroke patient rehabilitation proposed according to the embodiments of the present application can collect the cerebral blood oxygen signals of stroke patients through a portable multi-channel fNIRS device to construct a feature matrix, and construct a weighted brain network topology map based on at least one of the feature matrices to input into a deep learning model. Through a multi-task learning model and an adaptive feature fusion mechanism, temporal features, topological features, and manual features are dynamically integrated, and then a rehabilitation score and a trend prediction result can be output, thereby establishing an accurate tracking and detection rehabilitation evaluation and prediction method for stroke patient rehabilitation, and further realizing early intervention and personalized treatment during the rehabilitation process of stroke patients. Thus, in the related art, multi-channel fNIRS devices require more sensors and more complex settings, require patients to be tested in a fixed environment, cannot provide rehabilitation evaluation for patients anytime and anywhere, limit their application in home rehabilitation, and have certain limitations in the long-term rehabilitation monitoring of stroke patients and other problems are solved.

[0127] Figure 5 The structural schematic diagram of the electronic device provided by the embodiments of the present application. The electronic device may include:

[0128] A memory 501, a processor 502, and a computer program stored on the memory 501 and executable on the processor 502.

[0129] When the processor 502 executes the program, it implements the tracking and detection method for stroke patient rehabilitation provided in the above embodiments.

[0130] Further, the electronic device further includes:

[0131] A communication interface 503 for communication between the memory 501 and the processor 502.

[0132] The memory 501 is used to store a computer program executable on the processor 502.

[0133] The memory 501 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory.

[0134] If the memory 501, the processor 502, and the communication interface 503 are implemented independently, the communication interface 503, the memory 501, and the processor 502 can be interconnected through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 only a thick line is used to represent it in Figure 5 , but it does not mean that there is only one bus or one type of bus.

[0135] Optionally, in a specific implementation, if the memory 501, the processor 502, and the communication interface 503 are integrated on a single chip, the memory 501, the processor 502, and the communication interface 503 can communicate with each other through an internal interface.

[0136] The processor 502 may be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0137] The embodiments of the present application also provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned tracking detection method for stroke patient rehabilitation is implemented.

[0138] The embodiments of the present application also provide a computer program product, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned tracking detection method for stroke patient rehabilitation is implemented.

[0139] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc., mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of this application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0140] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "N" is at least two, such as two, three, etc., unless otherwise specifically defined.

[0141] Any process or method description shown in a flowchart or described in other ways herein may be understood to represent a module, segment, or portion of code including one or N executable instructions for implementing a customized logical function or process, and the scope of the preferred embodiments of this application includes additional implementations, where the functions may be executed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of this application pertain.

[0142] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection part (electronic device) having one or N wirings, a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.

[0143] It should be understood that various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above-described embodiments, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0144] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the methods of the above-described embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0145] In addition, each functional unit in various embodiments of the present application may be integrated into a processing module, or each unit may exist physically alone, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0146] The above-mentioned storage medium may be a read-only memory, a magnetic disk, an optical disc, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A tracking and detection method for stroke patient rehabilitation, characterized in that, Applied to the model construction stage, wherein the method includes the following steps: Obtain the cerebral blood oxygen signals of stroke patients in the recovery period collected by an oligolead near-infrared spectroscopy device; Preprocess and amplify the cerebral blood oxygen signals to obtain processed data, extract near-infrared signals of multiple target key frequencies from the processed data to construct a feature matrix, and construct a weighted brain network topology map based on at least one of the feature matrices; Merge the blood oxygen concentration data corresponding to the cerebral blood oxygen signals into a three-dimensional matrix and input it into a bidirectional long short-term memory network, and input the weighted brain network topology map into a multi-core graph convolutional network to extract temporal features and topological spatial features respectively, and map the temporal features, the topological spatial features and the manual features to the same dimension and then splice and fuse them, introduce a self-attention mechanism to dynamically adjust the feature weights, and construct a patient rehabilitation status and trend evaluation model for predicting the rehabilitation trend.

2. The method according to claim 1, wherein The preprocessing and amplification of the cerebral blood oxygen signals to obtain processed data, extracting near-infrared signals of multiple target key frequencies from the processed data to construct a feature matrix, and constructing a weighted brain network topology map based on the feature matrix includes: Convert the cerebral blood oxygen signals into the blood oxygen concentration data, and remove systematic chronic drift, remove jump noise, and remove irrelevant physiological noise and low-frequency noise to obtain initial processed data; Decompose and reconstruct the initial processed data based on wavelet transform to extract different frequency band sequences; Based on the different frequency band sequences, enhance the data by means of data segmentation and reconstruction to obtain the processed blood oxygen signals; Calculate various time-frequency features and frequency-specific functional connection features based on wavelet phase coherence for the processed blood oxygen signals to generate the feature matrix, and construct a weighted brain network topology map based on at least one of the feature matrices.

3. The method according to claim 1, characterized in that, The output layer of the patient rehabilitation status evaluation model includes a regression task branch and a classification task branch. Among them, the regression task branch is used to quantitatively evaluate the patient's rehabilitation status, the classification task branch is used to predict the patient's rehabilitation trend, the regression task branch uses the mean square error as the loss function, and the classification task branch uses the cross entropy as the loss function.

4. The method according to claim 3, wherein The construction of the patient rehabilitation status and trend evaluation model for predicting the rehabilitation trend includes: Optimize the joint loss function through a multi-task optimization strategy, and dynamically adjust the weight factors of the regression task and the classification task to balance the optimization priorities of the tasks.

5. The method according to claim 4, wherein Wherein, The optimization formula of the joint loss function is: Among them, L regression and L classification are the loss functions of the regression task and the classification task respectively, and σ reg and σ cia are the weight factors of the regression task and the classification task respectively; The adjustment formula of the weight factor is: where L regression and L classification are the loss functions of the regression task and the classification task respectively, ε0 is the initial smoothing term to avoid numerical explosion caused by a zero denominator, and γ is the decay coefficient.

6. A tracking and detection device for the rehabilitation of stroke patients, characterized in that, Applied to the model construction stage, wherein the device includes: An acquisition module for acquiring the cerebral blood oxygen signals of stroke patients in the recovery period collected by an oligolead near-infrared spectroscopy device; A processing module for preprocessing and amplifying the cerebral blood oxygen signals to obtain processed data, extracting near-infrared signals of multiple target key frequencies from the processed data to construct a feature matrix, and constructing a weighted brain network topology map based on at least one of the feature matrices; A prediction module, configured to merge the blood oxygen concentration data corresponding to the cerebral blood oxygen signal into a three-dimensional matrix, so as to input it into a bidirectional long short-term memory network, and input the weighted brain network topology map into a multi-core graph convolutional network, respectively extract temporal features and topological spatial features, and map the temporal features, the topological spatial features and manual features to a unified dimension and then splice and fuse them, introduce a self-attention mechanism to dynamically adjust the feature weights, and construct a patient rehabilitation status and trend evaluation model for predicting the rehabilitation trend.

7. The device according to claim 6, characterized in that, The processing module includes: A conversion unit, configured to convert the cerebral blood oxygen signal into the blood oxygen concentration data, and remove systematic chronic drift, remove jump noise, and remove irrelevant physiological noise and low-frequency noise, so as to obtain initial processed data; An extraction unit, configured to decompose and reconstruct the initial processed data based on wavelet transform, so as to extract different frequency band sequences; An enhancement unit, configured to enhance the data by using a data segmentation and reconstruction method based on the different frequency band sequences, so as to obtain a processed blood oxygen signal; A generation unit, configured to calculate a variety of time-frequency features and frequency-specific functional connection features based on wavelet phase coherence for the processed blood oxygen signal, generate the feature matrix, and construct a weighted brain network topology map based on at least one of the feature matrix.

8. An electronic device, characterized in that, It includes: A memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the program to implement the tracking detection method for stroke patient rehabilitation according to any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to be used for implementing the tracking detection method for stroke patient rehabilitation according to any one of claims 1-5.

10. A computer program product, comprising a computer program, characterized in that, The computer program is executed to be used for implementing the tracking detection method for stroke patient rehabilitation according to any one of claims 1-5.

Images