An elderly motion cognitive impairment personalized rehabilitation scheme recommendation system

Abstract

The application discloses a kind of old person motor cognitive impairment personalized rehabilitation scheme recommendation system.The system includes: data acquisition module: gather the case data about patient;Blood test data processing module: patient blood test data is embedded representation;Complaint text processing module: patient complaint text is embedded representation;Rehabilitation scheme acquisition module: single rehabilitation measure is represented, the representation of compound rehabilitation scheme is completed, rehabilitation measure is classified based on patient individualization, the representation of rehabilitation scheme is obtained by simulating the interaction of multiple rehabilitation measures;Rehabilitation scheme recommendation module: using the rehabilitation measure scoring model trained to complete old person motor cognitive impairment personalized rehabilitation scheme recommendation.The present application solves the problem that the convergence process is slow and even overfitting caused by the lack of embedding process of user and item in the current recommendation algorithm;And the problem of lacking consideration of patient individualization in single rehabilitation measure recommendation.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation medicine, specifically to a personalized rehabilitation program recommendation system for motor cognitive impairment in the elderly. Background Technology

[0002] With the development of medical informatization and big data technology, integrated platforms for assessment and intervention in rehabilitation medicine have gradually become a research hotspot in the medical field. Traditional rehabilitation methods mainly rely on the experience and professional knowledge of doctors. While this method is effective, it fails to fully consider the individual physical condition of patients, providing only general cognitive impairment treatment. The rehabilitation plan lacks specificity, resulting in poor rehabilitation efficiency and effectiveness (see "AI-based Capacity Assessment and Rehabilitation Treatment Recommendation System (CN202311358422.1) and Methods and Systems for Describing and Recommending Optimal Rehabilitation Plans in Adaptive Telemedicine (CN202110443973.2)"). Summary of the Invention

[0003] This invention primarily focuses on elderly individuals with cognitive impairment. The core of developing an integrated platform lies in utilizing patients' medical records and related information to provide personalized rehabilitation plans, thereby improving the accuracy and efficiency of rehabilitation. Existing research indicates that collaborative filtering (CF) and graph convolutional networks (GCNs) play significant roles in multi-objective combined recommendation.

[0004] To address the shortcomings of existing technologies, this invention combines classical recommendation algorithms with deep learning networks to provide a personalized recommendation system for rehabilitation programs for motor cognitive impairment in the elderly, based on patient blood test data and subjective complaints. This system employs collaborative filtering and graph convolutional networks. The system's rehabilitation program recommendations based on patient blood test data and subjective complaints have the following advantages: First, blood test data can be obtained through a single blood examination. This method is both quick and accurate, comprehensively reflecting the patient's health status and various physiological indicators. Blood test data includes, but is not limited to, key indicators such as glycated hemoglobin, low-density lipoprotein, and triglycerides, providing reliable foundational information for rehabilitation program recommendations. Second, patients with the same disease often exhibit different individualized physical signs in their blood test data. These individualized signs not only reflect the commonalities of the disease but also reveal individual differences among patients. Based on these individualized signs, more precise personalized rehabilitation recommendations can be achieved, thereby improving rehabilitation outcomes. Third, in addition to considering blood test data, the rehabilitation program recommendation also incorporates multimodal information, including patient subjective complaints, medical records, and other data sources. By integrating multimodal information, the system can gain a more comprehensive understanding of the patient's health status, thereby improving the accuracy and reliability of the recommendations. Finally, the rehabilitation recommendation system can comprehensively analyze the self-consistent relationship between compound rehabilitation exercises and the patient's physical condition, and predict the rehabilitation effects of different rehabilitation programs on individualized patients, avoiding potential adverse reactions. This rehabilitation treatment plan, which integrates multiple single rehabilitation exercises, makes rehabilitation program recommendations more scientific and safe, and shortens the patient's rehabilitation cycle.

[0005] This invention combines the advantages of collaborative filtering and graph convolutional networks to propose a personalized recommendation system for rehabilitation programs for elderly individuals with motor cognitive impairment. The system first embeds the patient's blood test data using an autoencoder and then embeds the patient's chief complaint text using a Long Short-Term Memory (LSTM) network incorporating attention mechanisms. Subsequently, a convolutional neural network is used to represent rehabilitation exercises. An initial graph convolutional network is used to simulate the interaction between rehabilitation exercises and the patient's personalized approach. Finally, the rehabilitation program representation is obtained through information aggregation and transmission. Finally, a rehabilitation treatment recommendation score is generated by combining collaborative filtering and a recommendation model with an MLP interaction layer.

[0006] This system can not only improve the accuracy and rationality of rehabilitation program recommendations, but also help doctors make more scientific decisions in certain complex situations, thereby improving the rehabilitation effect and efficiency of patients and reducing medical risks.

[0007] This invention addresses the recommendation problem of rehabilitation programs for motor cognitive impairment based on patient blood test data and subjective complaints. It collects patient case data, including blood test data, subjective complaint text, and historical rehabilitation programs as ground truth for training. An autoencoder and MSE loss function are used to embed the patient's blood test data. An LSTM network with an attention mechanism is used to embed the patient's subjective complaint text. Subsequently, one-hot encoding is used to represent a single rehabilitation program (a single rehabilitation exercise). Then, CNN-RRI-GCN (Convolutional Neural Networks-Rehabilitation and Rehabilitation Interaction-Graph Convolutional Networks) is used to complete the representation of rehabilitation programs (combinations of multiple single rehabilitation exercises). A collaborative filtering with an MLP interaction layer and a BPRLoss loss function are used to train and obtain a recommendation score for the rehabilitation treatment program. This solves the problems of slow convergence and even overfitting caused by the lack of embedding processes between users and items in current recommendation algorithms, as well as the lack of consideration for patient individualization in recommending single rehabilitation measures.

[0008] The objective of this invention is achieved by at least one of the following technical solutions.

[0009] A personalized rehabilitation program recommendation system for motor cognitive impairment in the elderly, specifically including:

[0010] Data acquisition module: Collects patient case data, including patient blood test data, chief complaint text, and historical rehabilitation plan;

[0011] Blood test data processing module: Employs an autoencoder and MSE loss function to embed and represent patient blood test data;

[0012] Chief complaint text processing module: Employs an LSTM network with an attention mechanism to embed and represent the patient's chief complaint text;

[0013] The rehabilitation plan acquisition module uses one-hot encoding to represent individual rehabilitation measures, and then uses a convolutional neural network (CNN-RRI-GCN) to represent complex rehabilitation plans. It also uses a convolutional neural network to predict rehabilitation measures for patient-specific event classification, and initializes a graph convolutional network to simulate the interaction of multiple rehabilitation measures. The rehabilitation plan representation is obtained through information aggregation, transmission, and global node pooling.

[0014] The rehabilitation program recommendation module uses collaborative filtering with an MLP interaction layer to construct a rehabilitation measure scoring model, then uses the BPRLoss loss function to train the trained rehabilitation measure scoring model, and finally uses the trained rehabilitation measure scoring model to recommend personalized rehabilitation programs for motor cognitive impairment in the elderly.

[0015] Furthermore, in the data processing module, for the purpose of applying blood test data research, the patients are those with heart disease, diabetes, hepatitis B, or the aforementioned comorbidities;

[0016] Blood test data include: glycated hemoglobin (Alc), low-density lipoprotein (Ldl), hepatitis B virus core antibody (Hbca), hepatitis B virus surface antibody (Hbsa), triglycerides (Tg), platelets (Plt), red blood cells (Rbc), cholesterol (Lym), hepatitis B e antigen (Hbeag), hepatitis B surface antigen (Hbsag), hepatitis B e antibody (Hbvea), and hemoglobin (Hbg).

[0017] Furthermore, in the blood test data processing module, an autoencoder is used to compress the patient's blood test data. The autoencoder includes an encoder and a decoder; both the encoder and decoder are three-layer MLPs. The blood test data is used as an input vector x and mapped to an embedding vector u1 through the encoder, as follows:

[0018] H e1 =g(W e1 ·x+b e1 (1)

[0019] H e2 =g(W e2 ·H e1 +b e2 (2)

[0020] u1 = W ep ·H e2 +b ep (3)

[0021] in, This represents a space for an n-dimensional column vector, meaning that the input vector x belongs to the n-dimensional real space. Let represent a space of p-dimensional column vectors, meaning the embedding vector u1 belongs to a p-dimensional real space; n represents the dimension of the input vector x, which is equal to the number of features in the blood test data; p is the dimension of the embedding vector u1, representing the dimension after data compression; g is the ReLU activation function; W e1 W e2 and W ep Both are weight matrices, W e1It is used to map the input vector x to the first hidden layer H in the encoder. e1 The weight matrix W e2 It is used to hide the first hidden layer H e1 Mapped to the second hidden layer H in the encoder e2 The weight matrix W ep It is used to hide the second hidden layer H in the encoder. e2 The weight matrix mapped to the embedding vector u1; b e1 b e2 and b ep These represent the bias vectors of the first, second, and third hidden layers in the autoencoder, respectively.

[0022] The embedding vector u1 is mapped back to the reconstructed input through the decoder. Specifically as follows:

[0023] H d1 =f(W d1 ·u1+b d1 (4)

[0024] H d2 =f(W d2 ·H d1 +b d2 (5)

[0025]

[0026] Where f is the ReLU activation function; W d1 W d2 and W dx Both are weight matrices, W d1 It is used to map the embedding vector u1 to the first decoding hidden layer H in the decoder. d1 The weight matrix W d2 It is used to decode the first hidden layer H d1 Mapped to the second decoding hidden layer H in the decoder d2 The weight matrix W dx It is used to hide the second decoding layer H d2 Mapping to the reconstructed output vector The weight matrix; b d1 b d2 and b dx These represent the bias vectors of the first, second, and third decoding hidden layers in the decoder, respectively.

[0027] Furthermore, in the blood test data processing module, the mean squared error (MSE) is used as the loss function to calculate the output vector reconstructed by the autoencoder. The difference between the input vector x and the loss function The expression is:

[0028]

[0029] Where, x i represents the i-th component in the input vector x, which represents the specific value of the i-th blood test feature in the blood test data; Represents the reconstructed output vector The i-th component is the data decoded by the autoencoder, and is related to the i-th component x in the input vector x. i Correspondingly;

[0030] Set convergence conditions, if If the value is less than 0.001, stop the iteration; otherwise, perform the following steps to update the parameters:

[0031]

[0032] m t =β1m t-1 +(1-β1)g t (9)

[0033]

[0034] Where t represents the iteration number, used to identify the current iteration step during the optimization process; g t m represents the gradient at the t-th iteration; t and v t Let W represent the estimated values ​​of the first and second moments at the t-th iteration, respectively; t and b t Let represent the weight matrix and bias vector at the t-th iteration, respectively; This represents the second-order moment estimate after bias correction;

[0035] After iterative calculation and meeting the convergence condition, the output u1 of the autoencoder is the embedded representation of the patient's blood test data.

[0036] Furthermore, in the chief complaint text processing module, the chief complaint text of the patient's dialogue with the doctor during the consultation is preprocessed by padding with empty characters to form a fixed-length text note; the chief complaint text note = (w1, w2, ..., w 200 Each word w in ) p Mapped to a vector Then the vectors y1, y2, y3, ..., y2 corresponding to the 200 words are... 200 A text matrix Y can be constructed to encode the entire text, thereby completing word embedding, as follows:

[0037] Word2 Vec(note)→Y (14)

[0038] The matrix Y is input into the LSTM network, and the cell state update calculation process is as follows:

[0039] f t =σ(W f ·[h t-1 y t ]+b f (15)

[0040] i t =σ(W p ·[h t-1 y t ]+b i (16)

[0041]

[0042] o t =σ(W o ·[h t-1 y t ]+b o (19)

[0043] h t =o t *tanh(C t (20)

[0044] Among them, W f W p W C and W o Both are weight matrices, W f W represents the weight matrix of the forget gate. p W represents the weight matrix of the input gate. C W represents the weight matrix of candidate memory cell states. o b represents the weight matrix of the output gate; f b i b C and b o Both are bias vectors, b f b represents the bias vector of the forget gate. i b represents the bias vector of the input gate. C b represents the bias vector of the candidate memory cell state. o h represents the bias vector of the output gate. t It is the hidden state of the LSTM network at time step t; C t It represents the cell state; σ is the sigmoid activation function; * represents the hardman product;

[0045] The hidden states of the above LSTM network form a matrix H = [h1, h2, ..., h...]. m ,...h 200 As input, calculate the weight matrix A, and the elements of weight matrix A. The weight matrix A represents the importance of the hidden state of the LSTM network at time step m to time step n; the weight matrix A is expressed in matrix form:

[0046] A = H·H T (twenty one)

[0047] Row normalization of the elements in the weight matrix A is performed using softmax:

[0048] A′=Softmax(A) (22)

[0049] Element-wise weighting; for each time step m, calculate the hidden state h of the LSTM network at that time step. m The corresponding row a in the normalized weight matrix A′ m Perform element weighting, for each hidden state h m It will be converted to a weighted hidden state:

[0050]

[0051] Represented using a matrix as V = A′H;

[0052] Finally, pooling is used to obtain the embedded representation u2 of the patient's chief complaint text:

[0053]

[0054] Among them, V n This represents the nth row of matrix V.

[0055] Furthermore, in the rehabilitation program acquisition module, the classification and prediction of RRI are studied based on the classification characteristics of rehabilitation measures; the classification of rehabilitation measures includes cognitive behavioral therapy, memory training, psychotherapy and gait training;

[0056] However, considering that the above-mentioned rehabilitation exercise classification features need to be represented by binary feature vectors with large dimensions, and the number of categories far exceeds the number of historical rehabilitation measures used in the cases, its sparsity and dimensionality are not suitable as the basis for representing rehabilitation measures.

[0057] In contrast, using one-hot encoding to represent rehabilitation measures is more reasonable, that is, using a vector d k To characterize a single rehabilitation measure, where d k = (0, 0, ..., 1, ..., 0), where only the k-th element is 1, and the rest are 0; dk This represents the one-hot code of the k-th rehabilitation measure, where the value of k ranges from 1 to N, and N is the number of different rehabilitation measures found from historical rehabilitation programs;

[0058] Using CNN to predict RRI: Combining the classification features of rehabilitation exercises, a similarity matrix is ​​calculated, and the corresponding two row vectors are selected to form a feature matrix for rehabilitation measure pairs, which is used as input to the CNN model; then, the probability of various RRI events is predicted by the CNN model, and the specific calculation method is as follows:

[0059] For one type of measure classification, let s be the denoting factor. a s b Let be the binary feature vectors of rehabilitation measures a and b, respectively, regarding cognitive behavioral therapy. This represents 116 evaluation dimensions related to cognitive behavior; the Jaccard similarity coefficient between these two rehabilitation measures at the cognitive behavioral therapy level is calculated, expressed as follows:

[0060]

[0061] Among them, M 11 In vector s a and s b In the expression, M represents the number of indices where the corresponding element value is 1. 01 In vector s a and s b In the table, the corresponding element values ​​are the number of indices of 0 and 1, respectively; M 10 In vector s a and s b In the table, the corresponding element values ​​are the number of indices of 1 and 0, respectively; the Jaccard similarity coefficient S a，b The value range is [0, 1];

[0062] Similarly, calculate the similarity coefficient between any two rehabilitation measures at the measure classification level. Let N represent the number of different rehabilitation measures found from historical rehabilitation programs. Then, construct an N×N similarity evaluation matrix S, where the element in the a-th row or b-th column of the similarity evaluation matrix S is S_a. a，b , represents the similarity coefficient between rehabilitation measures a and b at the cognitive behavioral therapy level, and S a，b =S b，a The element in the a-th row or a-th column of the similarity evaluation matrix S represents the similarity feature vector between the a-th rehabilitation measure and other rehabilitation measures at the classification level of that measure.

[0063] Following this method of constructing a similarity matrix, similarity matrices can be constructed between rehabilitation measures across all measure categories (cognitive behavioral therapy, memory training, psychotherapy, and gait training), denoted as the similarity matrix C for cognitive behavioral therapy, the similarity matrix M for memory training, the similarity matrix P for psychotherapy, and the similarity matrix G for gait training.

[0064] For any one of the rehabilitation measures in the set, let's say it's the a-th rehabilitation measure. It has a row vector in each of the four similarity matrices CM, P, and G. This row vector represents the similarity feature vector between the a-th rehabilitation measure and other rehabilitation measures at different measure classification levels. These four similarity feature vectors are concatenated column-wise to form a 4×N matrix I, where N is the number of different rehabilitation measures found from historical rehabilitation programs, and I is the comprehensive similarity matrix between the a-th rehabilitation measure and other rehabilitation measures at various levels.

[0065] A rehabilitation intervention interaction prediction network, comprising multiple convolutional and fully connected layers, is used to predict the interactions between rehabilitation interventions. A comprehensive similarity matrix I is input into the network to extract subspace features from the input tensor. The network consists of three fully connected layers. Based on the 65 categories of RRI events defined in the RRI-DMI (Rehabilitation and Rehabilitation Interaction-Dynamic Measure Index), and considering that the interaction between rehabilitation interventions (A, B) can be categorized into five types: A→B synergistic, B→A synergistic, no significant interaction, A→B antagonistic, and B→A antagonistic, where A and B correspond to the i-th and j-th rehabilitation interventions, respectively, the number of neurons in the three fully connected layers is set to 256, 65, and 5, respectively. The first two fully connected layers use linear activation functions, and the last layer uses a sigmoid activation function for normalized probability prediction. This network can predict the five types of interactions between rehabilitation interventions.

[0066] Furthermore, in the rehabilitation program acquisition module, the training of the rehabilitation measure interaction prediction network uses the Adam optimization algorithm, with an initial learning rate η = 0.001 and a loss function Loss of:

[0067]

[0068] Among them, y q The truth value of the RRI for the q-th interaction relationship: 0 indicates that there is no RRI, and 1 indicates that there is an RRI; p qThe predicted probabilities of the five interactions are represented, and the result with the highest probability is selected as the RRI prediction result.

[0069] The GCN edge weights are initialized based on the RRI events predicted using CNN to characterize the interactions and representations among multiple rehabilitation measures included in the rehabilitation program, as follows:

[0070] Take a case from the dataset, and denote the rehabilitation plan used in the case as R. Represent the rehabilitation plan R with a weighted graph. For each rehabilitation measure in the rehabilitation plan R, initialize a node representation in the graph. For any two rehabilitation measures in the rehabilitation plan R, if the RRI of the two rehabilitation measures exists according to the CNN prediction, then generate a directed edge between the corresponding two nodes in the graph structure, with a weight attached.

[0071]

[0072] Where v and u represent two different rehabilitation measures in rehabilitation program R, e vu This indicates the impact of rehabilitation measure v on u; considering rehabilitation measure d u and d v Furthermore, the effects and extent of these interactions vary depending on the individual patient. Therefore, a mask model is used to analyze the personalized impact of rehabilitation measures on patients.

[0073] This masking model is an MLP that receives a patient embedding vector u. The embedding vector is constructed by concatenating two feature vectors obtained from the blood test data processing module and the chief complaint text processing module. The concatenated vector u is used to personalize the patient's features. After processing, a vector with the same representation dimension as a single rehabilitation measure is output, called the mask vector. This mask vector is then element-wise multiplied with the rehabilitation measure representation to obtain... Specifically as follows:

[0074]

[0075] Wherein: This masked model MLP includes one hidden layer; the activation function used is the ReLU activation function, d u and d v These represent the initial representation vectors of rehabilitation measures u and v, respectively, obtained through one-hot encoding; and This represents the rehabilitation measures representation vector after adjustment using a patient-specific influence mask;

[0076] Calculated and Then, the data is concatenated and input into a pre-trained MLP model for further processing. and The concatenated vectors are then subjected to dimensionality reduction, linear mapping, and activation in sequence, resulting in an influence factor c used to update the edge weights. vu The details are as follows:

[0077]

[0078] Among them, c vu This represents the influencing factor, reflecting the degree of impact of the interaction between rehabilitation measures v and u on a specific patient;

[0079] The impact factor c obtained through the above steps vu This includes information on the extent of the interaction between two individual rehabilitation measures on a specific patient basis. This represents the updated edge weights, reflecting the strength of the interaction between rehabilitation measures after incorporating patient-specific factors; therefore, the edge weights can be updated accordingly.

[0080]

[0081] After initializing the nodes and edges, computation is performed using the GCN training method:

[0082] GCN learns node representations inductively by recursively aggregating and transforming the feature vectors of neighboring nodes; each layer update involves message passing, message aggregation, and node representation updates, and its mathematical expression is as follows:

[0083]

[0084] Here, v and u represent two different rehabilitation measures in the rehabilitation plan, and in the graph they represent two neighboring nodes, and there is a directed edge from u to v; This represents the message vector passed from node u to node v; This is the representation of node v at layer l in the GCN; N(v) represents the weight of the directed edge from u to v; N(v) represents the neighborhood of node v, from which node v collects information to update its aggregated message. Initialize to the corresponding rehabilitation measure characterization d u d v ;

[0085] The embedded representation g of the rehabilitation program is obtained using the following formula:

[0086]

[0087] Where σ is the ReLU activation function.

[0088] Furthermore, in the rehabilitation program recommendation module, the embedding results u1 and u2 obtained according to formulas (3) and (24) respectively are concatenated to obtain the complete embedding u representing the l-th patient. l and the characteristics of rehabilitation measures d k u l The represents the personalized embedding vector of the patient, which combines the embedding information of the patient's blood test data and chief complaint text. k The one-hot encoded embedding of the k-th rehabilitation measure is used to represent a single rehabilitation measure; the concatenated vector [u l |d k Mapping to values ​​through multiple fully connected layers Indicates based on the patient's personalized information u l As a measure of rehabilitation k Rating:

[0089] H e1 =g(W e1 ·[u l | d k ]+b e1 (35)

[0090] H e2 =g(W e2 ·H e1 +b e2 (36)

[0091] H e3 =g(W e3 ·H e2 +b e3 (37)

[0092]

[0093] Where g is the activation function; W e1 W e2 W e3 and W eo Both are weight matrices; b e1 b e2 b e3 b eo It is the bias vector; H e1 H e2 and H e3 These are the outputs of the hidden layer.

[0094] Furthermore, in the rehabilitation program recommendation module, an MLP model is used as the rehabilitation measure scoring model;

[0095] Using BprLoss as the loss function, the rehabilitation measure scoring model is trained. The mathematical expression of the loss function L1 is:

[0096]

[0097] Where: σ represents the sigmoid activation function; P n This represents the historical rehabilitation program used for the nth case in the dataset; j and k represent a single rehabilitation measure in the set of rehabilitation measures. For the L2 regularization used;

[0098] By minimizing the loss function, we can... As large as possible, that is, for the patient in the nth record, the corresponding rehabilitation measure d j Compared to rehabilitation measures with no history of use d k A higher score yields a rehabilitation measure scoring model that has completed initial training, an MLP model capable of personalized scoring of individual rehabilitation measures based on patient information.

[0099] The rehabilitation plan calculated by formula (34) is embedded into g, replacing the single rehabilitation measure representation d in formula (35), and fed into the rehabilitation measure scoring model after initial training. The loss function can be expressed as:

[0100]

[0101] During the update iteration, if L2(W, b) is less than 0.01, the iteration stops; otherwise, the following steps are performed to update the parameters:

[0102]

[0103] m t =β1m t-1 +(1-β1)g t (42)

[0104]

[0105] After iterative calculations and meeting the convergence condition, the final trained rehabilitation measure scoring model is obtained, which scores the rehabilitation plan based on the patient's individual needs.

[0106] H e1 =g(W1·[u i |g j ]+b1) (47)

[0107] H e2 =g(W2·H e1 +b2) (48)

[0108] H e3 =g(W3·H e2 +b3) (49)

[0109]

[0110] Furthermore, in the rehabilitation plan recommendation module, the recommendation process is as follows:

[0111] The final trained rehabilitation measure scoring model outputs a score for the rehabilitation plan based on the input patient embedding (including blood test data and chief complaint text) and historical rehabilitation plan embedding. This score reflects the patient's preference for different rehabilitation measures, thereby enabling personalized rehabilitation plan recommendations.

[0112] Recommending rehabilitation programs is crucial for alleviating the strain on medical resources and improving rehabilitation efficiency and quality. This invention combines the advantages of collaborative filtering and graph convolutional networks. It embeds blood test data of elderly patients with cognitive impairment using an autoencoder and embeds the patient's chief complaint text using an LSTM network with an attention mechanism. A convolutional neural network is used to represent individual rehabilitation measures. The interaction between rehabilitation measures is simulated by initializing a graph convolutional network. Finally, the representation of the rehabilitation program is obtained through information aggregation and transmission. Combining collaborative filtering and a recommendation model with an MLP interaction layer, a personalized score for the rehabilitation program is generated. This invention proposes a personalized recommendation method for rehabilitation programs for elderly patients with motor cognitive impairment based on patient blood test data and chief complaint information.

[0113] Compared with existing technical solutions, the advantages of the present invention are as follows: the combination recommendation of single rehabilitation measures based on patient blood test data and chief complaint has the following advantages: First, blood test data can be obtained through a single blood test; second, the blood test data of patients with the same disease reflect individual signs, thereby realizing personalized recommendations; third, considering multimodal information, the recommendation accuracy is high; fourth, the interaction between multiple rehabilitation measures can be analyzed.

[0114] To address the aforementioned challenges, this invention primarily focuses on the software component: it utilizes the MVC software architecture to build the software framework, and employs Qt and Python to develop the system's visual interface and the operational logic for each module, thus meeting the system's usage requirements.

[0115] (1) To embed patient information, a blood test data processing module and a chief complaint text processing module are proposed. The blood test data processing module learns a compressed representation of the patient's numerical data in an unsupervised manner by minimizing the reconstruction error, and uses an autoencoder and MSE loss function to embed the patient's blood test data. The chief complaint text processing module uses an LSTM network with an attention mechanism to embed the patient's chief complaint text. That is, it controls the information flow through a gating mechanism, captures the semantic relationships in the patient's chief complaint text, and focuses on the typical features of the patient's chief complaint text through an attention mechanism. In this way, it processes the complex structure of the patient's multi-dimensional information and captures its features to obtain the patient's embedded representation.

[0116] (2) To characterize a single rehabilitation measure, a rehabilitation scheme acquisition module based on a multi-layer convolutional network and a message aggregation mechanism is proposed. First, a single rehabilitation measure is binary encoded according to a binary descriptor, and the Jaccard similarity matrix is ​​calculated. Then, the feature vector is fed into a multi-layer convolutional network to capture the local features of the rehabilitation measure pair and predict the classification results of the interaction events of the rehabilitation measure pair. Finally, a graph convolutional network is initialized based on the results, and the complex interaction of the rehabilitation scheme is simulated through information transmission, node updates, and other processes. Finally, the node information is globally pooled to obtain the embedded representation of the rehabilitation scheme.

[0117] (3) Finally, a rehabilitation program recommendation module based on patient information is proposed. This model is based on collaborative filtering, which feeds the patient embedding and the representation of a single rehabilitation measure into a multilayer perceptron to realize high-order feature interaction and simulate the interaction effect between the patient and a single rehabilitation measure. Then, based on whether there is an interaction history with the individual patient in the dataset, the single rehabilitation measure is divided into positive and negative samples. The Bayesian personalized ranking loss function is calculated so that the personalized score of the positive sample single rehabilitation measure combination is higher than that of the negative sample. The trained multilayer perceptron can realize personalized recommendation of single rehabilitation measures. Finally, the patient embedding and rehabilitation program embedding are fed into the model for retraining to obtain a personalized rehabilitation program recommendation model based on patient information. Attached Figure Description

[0118] Figure 1 This is a flowchart illustrating the workflow of a personalized rehabilitation program recommendation system for motor cognitive impairment in the elderly, as described in an embodiment of the present invention.

[0119] Figure 2 This is a data processing structure diagram of the rehabilitation measure scoring model in an embodiment of the present invention;

[0120] Figure 3 This is a schematic diagram of the blood test data processing module in an embodiment of the present invention;

[0121] Figure 4 This is a schematic diagram of the structure of the main complaint text processing module in an embodiment of the present invention;

[0122] Figure 5 This is a flowchart of the rehabilitation plan acquisition module in an embodiment of the present invention. Detailed Implementation

[0123] To make the objectives, technical solutions, and advantages of the present invention clearer, the specific implementation of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0124] Example:

[0125] A personalized rehabilitation program recommendation system for motor cognitive impairment in the elderly, such as Figure 1 As shown, it specifically includes:

[0126] Data acquisition module: Collects patient case data, including patient blood test data, chief complaint text, and historical rehabilitation plan;

[0127] Blood test data processing module: Employs an autoencoder and MSE loss function to embed and represent patient blood test data;

[0128] Chief complaint text processing module: Employs an LSTM network with an attention mechanism to embed and represent the patient's chief complaint text;

[0129] The rehabilitation plan acquisition module uses one-hot encoding to represent individual rehabilitation measures, and then uses a convolutional neural network (CNN-RRI-GCN) to represent complex rehabilitation plans. It also uses a convolutional neural network to predict rehabilitation measures for patient-specific event classification, and initializes a graph convolutional network to simulate the interaction of multiple rehabilitation measures. The rehabilitation plan representation is obtained through information aggregation, transmission, and global node pooling.

[0130] Rehabilitation program recommendation module: such as Figure 2 As shown, a rehabilitation measure scoring model is constructed using collaborative filtering with an MLP interaction layer. Then, the trained rehabilitation measure scoring model is obtained by training with the BPRLoss loss function. Finally, the trained rehabilitation measure scoring model is used to recommend personalized rehabilitation programs for motor cognitive impairment in the elderly.

[0131] Furthermore, in the data processing module, for the purpose of applying blood test data research, the patients are those with heart disease, diabetes, hepatitis B, or the aforementioned comorbidities;

[0132] Blood test data include: glycated hemoglobin (Alc), low-density lipoprotein (Ldl), hepatitis B virus core antibody (Hbca), hepatitis B virus surface antibody (Hbsa), triglycerides (Tg), platelets (Plt), red blood cells (Rbc), cholesterol (Lym), hepatitis B e antigen (Hbeag), hepatitis B surface antigen (Hbsag), hepatitis B e antibody (Hbvea), and hemoglobin (Hbg).

[0133] Furthermore, in the blood test data processing module, such as Figure 3 As shown, an autoencoder is used to compress patient blood test data. The autoencoder includes an encoder and a decoder; both the encoder and decoder are three-layer MLPs. The blood test data is used as an input vector x and mapped to an embedding vector u1 through the encoder, as follows:

[0134] H e1 =g(W e1 ·x+b e1 (1)

[0135] H e2 =g(W e2 ·H e1 +b e2 (2)

[0136] u1 = W ep ·H e2 +b ep (3)

[0137] in, This represents a space for an n-dimensional column vector, meaning that the input vector x belongs to the n-dimensional real space. Let represent a space of p-dimensional column vectors, meaning the embedding vector u1 belongs to a p-dimensional real space; n represents the dimension of the input vector x, which is equal to the number of features in the blood test data; p is the dimension of the embedding vector u1, representing the dimension after data compression; g is the ReLU activation function; W e1 W e2 and W ep Both are weight matrices, W e1 It is used to map the input vector x to the first hidden layer H in the encoder. e1 The weight matrix W e2 It is used to hide the first hidden layer H e1 Mapped to the second hidden layer H in the encoder e2 The weight matrix W ep It is used to hide the second hidden layer H in the encoder. e2 The weight matrix mapped to the embedding vector u1; b e1 b e2 and b epThese represent the bias vectors of the first, second, and third hidden layers in the autoencoder, respectively.

[0138] The embedding vector u1 is mapped back to the reconstructed input through the decoder. Specifically as follows:

[0139] H d1 =f(W d1 ·u1+b d1 (4)

[0140] H d2 =f(W d2 ·H d1 +b d2 (5)

[0141]

[0142] Where f is the ReLU activation function; W d1 W d2 and W dx Both are weight matrices, W d1 It is used to map the embedding vector u1 to the first decoding hidden layer H in the decoder. d1 The weight matrix W d2 It is used to decode the first hidden layer H d1 Mapped to the second decoding hidden layer H in the decoder d2 The weight matrix W dx It is used to hide the second decoding layer H d2 Mapping to the reconstructed output vector The weight matrix; b d1 b d2 and b dx These represent the bias vectors of the first, second, and third decoding hidden layers in the decoder, respectively.

[0143] In one embodiment, the blood test data processing module uses mean squared error (MSE) as the loss function to calculate the output vector reconstructed by the autoencoder. The difference between the input vector x and the loss function The expression is:

[0144]

[0145] Where, x i represents the i-th component in the input vector x, which represents the specific value of the i-th blood test feature in the blood test data; Represents the reconstructed output vector The i-th component is the data decoded by the autoencoder, and is related to the i-th component x in the input vector x. iCorrespondingly;

[0146] Set convergence conditions, if If the value is less than 0.001, stop the iteration; otherwise, perform the following steps to update the parameters:

[0147]

[0148] m t =β1m t-1 +(1-β1)g t (9)

[0149]

[0150] Where t represents the iteration number, used to identify the current iteration step during the optimization process; g t m represents the gradient at the t-th iteration; t and v t Let W represent the estimated values ​​of the first and second moments at the t-th iteration, respectively; t and b t Let represent the weight matrix and bias vector at the t-th iteration, respectively; This represents the second-order moment estimate after bias correction;

[0151] After iterative calculation and meeting the convergence condition, the output u1 of the autoencoder is the embedded representation of the patient's blood test data.

[0152] In one embodiment, such as Figure 4 As shown, in the chief complaint text processing module, the chief complaint text of the patient's dialogue with the doctor during the consultation is preprocessed by padding with empty characters to form a fixed-length text note of length 200; Word2Vec is used to convert the chief complaint text note = (w1, w2, ..., w 200 Each word w in ) p Mapped to a vector Then the vectors y1, y2, y3, ..., y2 corresponding to the 200 words are... 200 A text matrix Y can be constructed to encode the entire text, thereby completing word embedding, as follows:

[0153] Word2Vec(note)→Y (14)

[0154] The matrix Y is input into the LSTM network, and the cell state update calculation process is as follows:

[0155] f t =σ(W f ·[h t-1 y t ]+b f (15)

[0156] i t =σ(W p ·[h t-1 y t ]+b i (16)

[0157]

[0158] o t =σ(W o ·[h t-1 y t ]+b o (19)

[0159] h t =o t *tanh(C t (20)

[0160] Among them, W f W p W C and W o Both are weight matrices, W f W represents the weight matrix of the forget gate. p W represents the weight matrix of the input gate. C W represents the weight matrix of candidate memory cell states. o b represents the weight matrix of the output gate; f b i b C and b o Both are bias vectors, b f b represents the bias vector of the forget gate. i b represents the bias vector of the input gate. C b represents the bias vector of the candidate memory cell state. o h represents the bias vector of the output gate. t It is the hidden state of the LSTM network at time step t; C t It represents the cell state; σ is the sigmoid activation function; * represents the hardman product;

[0161] The hidden states of the above LSTM network form a matrix H = [h1, h2, ..., h...]. m ,...h 200 As input, calculate the weight matrix A, and the elements of weight matrix A. The weight matrix A represents the importance of the hidden state of the LSTM network at time step m to time step n; the weight matrix A is expressed in matrix form:

[0162] A = H·H T(twenty one)

[0163] Row normalization of the elements in the weight matrix A is performed using softmax:

[0164] A′=Softmax(A) (22)

[0165] Element-wise weighting; for each time step m, calculate the hidden state h of the LSTM network at that time step. m The corresponding row a in the normalized weight matrix A′ m Perform element weighting, for each hidden state h m It will be converted to a weighted hidden state:

[0166]

[0167] Represented using a matrix as V = A′H;

[0168] Finally, pooling is used to obtain the embedded representation u2 of the patient's chief complaint text:

[0169]

[0170] Among them, V n This represents the nth row of matrix V.

[0171] In one embodiment, such as Figure 5 As shown, in the rehabilitation program acquisition module, the classification and prediction of RRI are studied based on the classification characteristics of rehabilitation measures; the classification of rehabilitation measures includes cognitive behavioral therapy, memory training, psychotherapy and gait training.

[0172] However, considering that the above-mentioned rehabilitation exercise classification features need to be represented by binary feature vectors with large dimensions, and the number of categories far exceeds the number of historical rehabilitation measures used in the cases, its sparsity and dimensionality are not suitable as the basis for representing rehabilitation measures.

[0173] In contrast, using one-hot encoding to represent rehabilitation measures is more reasonable, that is, using a vector d k To characterize a single rehabilitation measure, where d k = (0, 0, ..., 1, ..., 0), where only the k-th element is 1, and the rest are 0; d k This represents the one-hot code of the k-th rehabilitation measure, where the value of k ranges from 1 to N, and N is the number of different rehabilitation measures found from historical rehabilitation programs;

[0174] Using CNN to predict RRI: Combining the classification features of rehabilitation exercises, a similarity matrix is ​​calculated, and the corresponding two row vectors are selected to form a feature matrix for rehabilitation measure pairs, which is used as input to the CNN model; then, the probability of various RRI events is predicted by the CNN model, and the specific calculation method is as follows:

[0175] For one type of measure classification, let s be the denoting factor. a s b Let be the binary feature vectors of rehabilitation measures a and b, respectively, regarding cognitive behavioral therapy. This represents 116 evaluation dimensions related to cognitive behavior; the Jaccard similarity coefficient between these two rehabilitation measures at the cognitive behavioral therapy level is calculated, expressed as follows:

[0176]

[0177] Among them, M 11 In vector s a and s b In the expression, M represents the number of indices where the corresponding element value is 1. 01 In vector s a and s b In the table, the corresponding element values ​​are the number of indices of 0 and 1, respectively; M 10 In vector s a and s b In the table, the corresponding element values ​​are the number of indices of 1 and 0, respectively; the Jaccard similarity coefficient S a，b The value range is [0, 1];

[0178] Similarly, calculate the similarity coefficient between any two rehabilitation measures at the measure classification level. Let N represent the number of different rehabilitation measures found from historical rehabilitation programs. Then, construct an N×N similarity evaluation matrix S, where the element in the a-th row or b-th column of the similarity evaluation matrix S is S_a. a，b , represents the similarity coefficient between rehabilitation measures a and b at the cognitive behavioral therapy level, and S a，b =S b，a The element in the a-th row or a-th column of the similarity evaluation matrix S represents the similarity feature vector between the a-th rehabilitation measure and other rehabilitation measures at the classification level of that measure.

[0179] Following this method of constructing a similarity matrix, similarity matrices can be constructed between rehabilitation measures across all measure categories (cognitive behavioral therapy, memory training, psychotherapy, and gait training), denoted as the similarity matrix C for cognitive behavioral therapy, the similarity matrix M for memory training, the similarity matrix P for psychotherapy, and the similarity matrix G for gait training.

[0180] For any one of the rehabilitation measures in the set, let's say it's the a-th rehabilitation measure. It has a row vector in each of the four similarity matrices C, M, P, and G. This row vector represents the similarity feature vector between the a-th rehabilitation measure and other rehabilitation measures at different measure classification levels. These four similarity feature vectors are concatenated column-wise to form a 4×N matrix I, where N is the number of different rehabilitation measures found from historical rehabilitation programs, and I is the comprehensive similarity matrix between the a-th rehabilitation measure and other rehabilitation measures at various levels.

[0181] In one embodiment, a rehabilitation intervention interaction prediction network, comprising multiple convolutional and fully connected layers, is used to predict the interactions between rehabilitation interventions. A comprehensive similarity matrix I is input into the network to extract subspace features from the input tensor. The network consists of three fully connected layers. Based on the 65 categories of RRI events defined in the RRI-DMI (Rehabilitation and Rehabilitation Interaction-Dynamic Measure Index), and considering that the interaction between rehabilitation interventions (A, B) can be categorized into five types: A→B synergistic, B→A synergistic, no significant interaction, A→B antagonistic, and B→A antagonistic, where A and B correspond to the i-th and j-th rehabilitation interventions, respectively, the number of neurons in the three fully connected layers is set to 256, 65, and 5, respectively. The first two fully connected layers use linear activation functions, and the last layer uses a sigmoid activation function for normalized probability prediction. The rehabilitation intervention interaction prediction network can predict these five types of interactions between rehabilitation interventions.

[0182] Furthermore, in the rehabilitation program acquisition module, the training of the rehabilitation measure interaction prediction network uses the Adam optimization algorithm, with an initial learning rate η = 0.001 and a loss function Loss of:

[0183]

[0184] Among them, y q The truth value of the RRI for the q-th interaction relationship: 0 indicates that there is no RRI, and 1 indicates that there is an RRI; p q The predicted probabilities of the five interactions are represented, and the result with the highest probability is selected as the RRI prediction result.

[0185] The GCN edge weights are initialized based on the RRI events predicted using CNN to characterize the interactions and representations among multiple rehabilitation measures included in the rehabilitation program, as follows:

[0186] Take a case from the dataset, and denote the rehabilitation plan used in the case as R. Represent the rehabilitation plan R with a weighted graph. For each rehabilitation measure in the rehabilitation plan R, initialize a node representation in the graph. For any two rehabilitation measures in the rehabilitation plan R, if the RRI of the two rehabilitation measures exists according to the CNN prediction, then generate a directed edge between the corresponding two nodes in the graph structure, with a weight attached.

[0187]

[0188] Where v and u represent two different rehabilitation measures in rehabilitation program R, e vu This indicates the impact of rehabilitation measure v on u; considering rehabilitation measure d u and d v Furthermore, the effects and extent of these interactions vary depending on the individual patient. Therefore, a mask model is used to analyze the personalized impact of rehabilitation measures on patients.

[0189] This masking model is an MLP that receives a patient embedding vector u. The embedding vector is constructed by concatenating two feature vectors obtained from the blood test data processing module and the chief complaint text processing module. The concatenated vector u is used to personalize the patient's features. After processing, a vector with the same representation dimension as a single rehabilitation measure is output, called the mask vector. This mask vector is then element-wise multiplied with the rehabilitation measure representation to obtain... Specifically as follows:

[0190]

[0191] Wherein: This masked model MLP includes one hidden layer; the activation function used is the ReLU activation function, d u and d v These represent the initial representation vectors of rehabilitation measures u and v, respectively, obtained through one-hot encoding; and This represents the rehabilitation measures representation vector after adjustment using a patient-specific influence mask;

[0192] Calculated and Then, the data is concatenated and input into a pre-trained MLP model for further processing. and The concatenated vectors are then subjected to dimensionality reduction, linear mapping, and activation in sequence, resulting in an influence factor c used to update the edge weights. vu The details are as follows:

[0193]

[0194] Among them, cvu This represents the influencing factor, reflecting the degree of impact of the interaction between rehabilitation measures v and u on a specific patient;

[0195] The impact factor c obtained through the above steps vu This includes information on the extent of the interaction between two individual rehabilitation measures on a specific patient basis. This represents the updated edge weights, reflecting the strength of the interaction between rehabilitation measures after incorporating patient-specific factors; therefore, the edge weights can be updated accordingly.

[0196]

[0197] After initializing the nodes and edges, computation is performed using the GCN training method:

[0198] GCN learns node representations inductively by recursively aggregating and transforming the feature vectors of neighboring nodes; each layer update involves message passing, message aggregation, and node representation updates, and its mathematical expression is as follows:

[0199]

[0200] Here, v and u represent two different rehabilitation measures in the rehabilitation plan, and in the graph they represent two neighboring nodes, and there is a directed edge from u to v; This represents the message vector passed from node u to node v; This is the representation of node v at layer l in the GCN; N(v) represents the weight of the directed edge from u to v; N(v) represents the neighborhood of node v, from which node v collects information to update its aggregated message. Initialize to the corresponding rehabilitation measure characterization d u d v ;

[0201] The embedded representation g of the rehabilitation program is obtained using the following formula:

[0202]

[0203] Where σ is the ReLU activation function.

[0204] In one embodiment, in the rehabilitation program recommendation module, the embedding results u1 and u2 obtained according to formulas (3) and (24) respectively are concatenated to obtain the complete embedding u representing the l-th patient. l and the characteristics of rehabilitation measures d k u l d represents the personalized embedding vector of the l-th patient, which combines the embedding information of the patient's blood test data and chief complaint text. kThe one-hot encoded embedding of the k-th rehabilitation measure is used to represent a single rehabilitation measure; the concatenated vector [u l |d k Mapping to values ​​through multiple fully connected layers Indicates based on the patient's personalized information u l As a measure of rehabilitation k Rating:

[0205] H e1 =g(W e1 ·[u l |d k ]+b e1 (35)

[0206] H e2 =g(W e2 ·H e1 +b e2 (36)

[0207] H e3 =g(W e3 ·H e2 +b e3 (37)

[0208]

[0209] Where g is the activation function; W e1 W e2 W e3 and W eo Both are weight matrices; b e1 b e2 b e3 b eo It is the bias vector; H e1 H e2 and H e3 These are the outputs of the hidden layer.

[0210] Furthermore, in the rehabilitation plan recommendation module, such as Figure 2 As shown, an MLP model is used as a scoring model for rehabilitation measures;

[0211] Using BprLoss as the loss function, the rehabilitation measure scoring model is trained. The mathematical expression of the loss function L1 is:

[0212]

[0213] Where: σ represents the sigmoid activation function; P n This represents the historical rehabilitation program used for the nth case in the dataset; j and k represent a single rehabilitation measure in the set of rehabilitation measures. For the L2 regularization used;

[0214] By minimizing the loss function, we can... As large as possible, that is, for the patient in the nth record, the corresponding rehabilitation measure d j Compared to rehabilitation measures with no history of use d k A higher score yields a rehabilitation measure scoring model that has completed initial training, an MLP model capable of personalized scoring of individual rehabilitation measures based on patient information.

[0215] The rehabilitation plan calculated by formula (34) is embedded into g, replacing the single rehabilitation measure representation d in formula (35), and fed into the rehabilitation measure scoring model after initial training. The loss function can be expressed as:

[0216]

[0217] During the update iteration, if L2(W, b) is less than 0.01, the iteration stops; otherwise, the following steps are performed to update the parameters:

[0218]

[0219] m t =β1m t-1 +(1-β1)g t (42)

[0220]

[0221]

[0222] After iterative calculations and meeting the convergence condition, the final trained rehabilitation measure scoring model is obtained, which scores the rehabilitation plan based on the patient's individual needs.

[0223] H e1 =g(W1·[u i |g j ]+b1) (47)

[0224] H e2 =g(W2·H e1 +b2) (48)

[0225] H e3 =g(W3·H e2 +b3) (49)

[0226]

[0227] In one embodiment, the recommendation process in the rehabilitation plan recommendation module is as follows:

[0228] The final trained rehabilitation measure scoring model outputs a score for the rehabilitation plan based on the input patient embedding (including blood test data and chief complaint text) and historical rehabilitation plan embedding. This score reflects the patient's preference for different rehabilitation measures, thereby enabling personalized rehabilitation plan recommendations.

Claims

1. A personalized rehabilitation program recommendation system for motor cognitive impairment in the elderly, characterized in that, Specifically, it includes: Data acquisition module: Collects patient case data, including patient blood test data, chief complaint text, and historical rehabilitation plan; Blood test data processing module: Employs an autoencoder and MSE loss function to embed and represent patient blood test data; Chief complaint text processing module: Employs an LSTM network with an attention mechanism to embed and represent the patient's chief complaint text; The rehabilitation plan acquisition module employs one-hot encoding to represent individual rehabilitation measures, which are obtained from historical rehabilitation plans. A CNN-based interaction prediction network predicts the relative importance (RRI) between rehabilitation measures. Each rehabilitation measure is initialized with a node representation in the GCN. For any two rehabilitation measures in a rehabilitation plan R, if the CNN-based interaction prediction network predicts the existence of the RRI between these two measures, a directed edge with weights is generated between the corresponding two nodes in the graph structure. A mask model is used to analyze the personalized impact of rehabilitation measures on patients. The edge weights are updated based on the degree of influence of the interaction between two rehabilitation measures on specific individual patients. The mask model is an MLP that receives a patient embedding vector u. The embedding vector is constructed by concatenating two feature vectors obtained from the blood test data processing module and the chief complaint text processing module. The concatenated vector u is used to personalize the patient's features. Based on this, a graph convolutional network is initialized to simulate the interaction of multiple rehabilitation measures. The embedding representation of the rehabilitation plan is obtained through information aggregation, transmission and global node pooling. The rehabilitation program recommendation module: A rehabilitation measure scoring model is constructed using collaborative filtering with an MLP interaction layer. Then, the trained rehabilitation measure scoring model is obtained by using the BPRLoss loss function. Finally, the trained rehabilitation measure scoring model is used to recommend personalized rehabilitation programs for motor cognitive impairment in the elderly. The specific recommendation process is as follows: The final trained rehabilitation measure scoring model outputs a score for the rehabilitation plan based on the input patient embedding and rehabilitation plan embedding. This score reflects the patient's preference for different rehabilitation measures, thereby enabling personalized rehabilitation program recommendations. The patient embedding is composed of blood test data embedding and chief complaint text embedding.

2. The personalized rehabilitation program recommendation system for motor cognitive impairment in the elderly according to claim 1, characterized in that, In the data processing module, the patients are those with heart disease, diabetes, hepatitis B, or any of the above-mentioned comorbidities. Blood test data include: glycated hemoglobin, low-density lipoprotein, hepatitis B virus core antibody, hepatitis B virus surface antibody, triglycerides, platelets, red blood cells, cholesterol, hepatitis B e antigen, hepatitis B surface antigen, hepatitis B e antibody, and hemoglobin.

3. The personalized rehabilitation program recommendation system for motor cognitive impairment in the elderly according to claim 1, characterized in that, In the blood test data processing module, an autoencoder is used to compress the patient's blood test data. The autoencoder includes an encoder and a decoder; both the encoder and decoder are three-layer MLPs. The blood test data is used as an input vector x and mapped to an embedding vector u1 through the encoder, as follows: H e1 =g(W e1 ·x+b e1 ) (1) H e2 =g(W e2 ·H e1 +b e2 ) (2) u1=W ep ·H e2 +b ep (3) in, This represents a space for an n-dimensional column vector, meaning that the input vector x belongs to the n-dimensional real space. Let represent a space of p-dimensional column vectors, meaning the embedding vector u1 belongs to a p-dimensional real space; n represents the dimension of the input vector x, which is equal to the number of features in the blood test data; p is the dimension of the embedding vector u1, representing the dimension after data compression; g is the ReLU activation function; W e1 W e2 and W ep Both are weight matrices, W e1 It is used to map the input vector x to the first hidden layer H in the encoder. e1 The weight matrix W e2 It is used to hide the first hidden layer H e1 Mapped to the second hidden layer H in the encoder e2 The weight matrix W ep It is used to hide the second hidden layer H in the encoder. e2 The weight matrix mapped to the embedding vector u1; b e1 b e2 and b ep These represent the bias vectors of the first, second, and third hidden layers in the autoencoder, respectively. The embedding vector u1 is mapped back to the reconstructed input through the decoder. Specifically as follows: H d1 =f(W d1 ·u1+b d1 ) (4) H d2 =f(W d2 ·H d1 +b d2 ) (5) Where f is the ReLU activation function; W d1 W d2 and W dx Both are weight matrices, W d1 It is used to map the embedding vector u1 to the first decoding hidden layer H in the decoder. d1 The weight matrix W d2 It is used to decode the first hidden layer H d1 Mapped to the second decoding hidden layer H in the decoder d2 The weight matrix W dx It is used to hide the second decoding layer H d2 Mapping to the reconstructed output vector The weight matrix; b d1 b d2 and b dx These represent the bias vectors of the first, second, and third decoding hidden layers in the decoder, respectively.

Images