A method and system for exercise training recommendation for a population at risk of disability

Abstract

The application discloses a kind of exercise training recommendation method and system for risk population of disability.The method generates character image and obtains static characteristics by building the multi-level image system of patient, fusing the multidimensional data such as disease type, clinical characteristics, demographic information, exercise behavior characteristics of patient.The first exercise training scheme is pushed based on static characteristics, and the structured data, text data and image data of patient during training are collected, and the subsequent exercise training scheme is generated by collaborative filtering model, and the dynamic characteristics of patient are iteratively optimized using recurrent neural network model.The application can dynamically adjust exercise training scheme according to individual characteristics and training feedback of patient, improve the compliance and training effect of patient.

Description

TECHNICAL FIELD

[0001] The present application relates to a movement training recommendation method for a disability risk population, and also relates to a corresponding movement training recommendation system, and belongs to the technical field of medical care information science. BACKGROUND

[0002] According to the dual-coding theory in the cognitive theory, it is generally believed that humans have two main information processing systems: verbal system and non-verbal system. The two systems have different encoding methods. Studies have shown that the brain's memory effect and memory speed for visual materials are better than semantic memory, which indicates that the non-verbal system may have an advantage in processing images and spatial information. Although the two systems are independent in function, they can also activate and influence each other.

[0003] There is a theory in the field of cognition called constructivism, which holds that knowledge is constructed through the interaction of individuals with the environment. Through communication and cooperation with others, individuals can build and revise their understanding; when individuals encounter new information that conflicts with existing knowledge, cognitive conflict occurs. This conflict prompts individuals to re-evaluate and adjust their cognitive structure to accommodate new information. Moreover, knowledge construction is a continuous process. As individuals gain experience and develop cognitive abilities, their knowledge structure also evolves. Knowledge is constructed in specific situations, so learning in actual situations can better promote the understanding and application of knowledge

[0004] However, there is no movement training method that integrates information from the verbal system and the non-verbal system in the current treatment of disability risk populations. Moreover, for disability risk populations, individualized and targeted movement training programs need to be developed to improve or improve the decline in their cognitive, emotional, motor, verbal, and other abilities. SUMMARY

[0005] The primary technical problem to be solved by the present application is to provide a movement training recommendation method for a disability risk population.

[0006] Another technical problem to be solved by the present application is to provide a movement training recommendation system for a disability risk population.

[0007] To achieve the above technical purposes, the present application adopts the following technical solutions:

[0008] According to the first aspect of the embodiment of the present application, a movement training recommendation method for a disability risk population is provided, comprising the following steps:

[0009] Step one: obtain the disease type and clinical characteristics of the patient, demographic information, exercise research information and exercise behavior characteristics to fuse and construct a role image and obtain static characteristics; wherein the patient is a disability risk population;

[0010] Step two: generate and push the first exercise training program based on the role image;

[0011] Step three: collect structured data, text data and image data during exercise training;

[0012] Step four: input the data obtained in step three into a collaborative filtering model to generate and push the next exercise training program in combination with the role image;

[0013] Step five: collect structured data, text data and image data during exercise training in the manner of step three to obtain exercise training cycle data;

[0014] Step six: determine whether the exercise training program is the Nth, if so, go to step seven; if not, return to step four to generate the next exercise training program according to the characteristics in step five, N is a positive integer;

[0015] Step seven: based on the data collected in steps three and five, input the recurrent neural network model to iteratively optimize the exercise training program; wherein the recurrent neural network model introduces two LSTM networks; at each time step, one of the LSTM networks combines the dynamic feature vector of the patient at the previous time step and the static feature vector at the current time step to update the dynamic feature vector of the patient; the other LSTM network combines the dynamic feature vector of the exercise at the previous time step and the static feature vector at the current time step to update the dynamic feature vector of the exercise.

[0016] Preferably, the recurrent neural network model is optimized using the following function:

[0017]

[0018] Wherein θ represents the parameters to be learned; is a set of observed tuples in the training set, the tuple including the patient, the exercise training program and the time sequence; r ij|t represents the score of patient i at time t for exercise j; is the predicted value of r ij|t ; R represents a regularization function.

[0019] Preferably, the recurrent neural network model adopts an alternating subspace descent strategy; wherein, assuming that the motion state is fixed, the gradient is not propagated to the motion training program sequence, and the parameters of the patient sequence are updated by back-propagating the gradient of all patient scores; then, alternating between updating the patient sequence and updating the motion training program sequence.

[0020] Preferably, in step four, multi-modal data fusion is realized based on the structured data, the text data, and the image data to obtain user latent vectors and item latent vectors, which are input into an attribute fusion convolutional layer to recommend the next motion training program.

[0021] Preferably, step four includes the following sub-steps:

[0022] 1) Determine the key variables as one-dimensional data, two-dimensional data, and three-dimensional data;

[0023] 2) Create a data matrix according to the key variables;

[0024] 3) Add lag features as four-dimensional data;

[0025] 4) Convert to a feature vector using embedding technology;

[0026] 5) Generate item latent vectors and user latent vectors based on the feature vector for collaborative filtering learning to recommend a motion training program.

[0027] Preferably, in sub-step 4, the entity and the relationship are converted into vectors using the Bayesian TransR knowledge graph embedding method; the text data is converted into a text vector using the Bayesian sparse autoencoder embedding method; and the features of the image are extracted and converted into an image vector using the Bayesian sparse convolutional autoencoder.

[0028] Preferably, the multi-modal features and role image features extracted for the patient are used as independent variables, and the doctor's clinical motion prescription suggestion is used as a dependent variable, which are input into the collaborative filtering model; wherein, the loss function of the collaborative filtering model is a cosine contrast loss, which is used to maximize the similarity between positive sample pairs and minimize the similarity of negative sample pairs under margin constraints.

[0029] Preferably, the motion training recommendation method further includes:

[0030] According to the data in the last motion training program, the training effect is evaluated, and if the effect is improved, a motion training report is made and output or the process is directly ended; if the effect is not improved, the process returns to step seven to generate a new motion training program.

[0031] According to a second aspect of the embodiments of the present application, a movement training recommendation system for a risk population of disability is provided, comprising a processor and a memory, wherein the memory is coupled to the processor and is used to store one or more programs, when the programs are executed by the processor, the processor implements the movement training recommendation method for the risk population of disability as described above.

[0032] Compared with the prior art, the present application has the following technical advantages by formulating an individualized and targeted movement training scheme:

[0033] 1) A movement recommendation framework combining the advantages of multiple model algorithms is proposed. On the one hand, the framework combines collaborative filtering and auto-encoding learning technology to build a fusion analysis system for multi-modal data, which can effectively integrate multi-source information into the recommendation model, thereby enriching the analysis source and providing more comprehensive data support for movement recommendation. On the other hand, the recurrent neural network (RNN) is introduced to effectively calculate the time series data. Through the retrospective analysis of the movement data of patients at different time periods, the coherence and sustainability of the movement recommendation are ensured, thereby improving the patient's compliance.

[0034] 2) A training cycle data system is constructed, which divides the movement training process into annual cycles, large cycles, medium cycles and small cycles, and generates corresponding time series data sets for different cycles. When these data are introduced into the model for movement recommendation calculation, the time series dimension feature is increased. This enables the model to more fully and regularly utilize historical movement data, thereby assisting in generating a periodic movement recommendation scheme.

[0035] 3) A movement push system based on deep learning is constructed. The system integrates multi-dimensional movement training push logic into a complete process, which can continuously update the movement push content based on the patient's own multi-dimensional movement data. In this way, it ensures that the training scheme always matches the current state of the patient, thereby improving the training effect.

[0036] 4) A multi-level patient portrait system is constructed. The system gradually and deeply deconstructs and evaluates the patient's attribute features, behavior features and comprehensive features (including cognition, emotion, movement, speech, etc.), ensuring that the individual tag system has a detailed granularity. This provides a large amount of effective sequence information for subsequent input of patient data into the model for precise movement recommendation, further improving the accuracy and individualization of the recommendation. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 For the first embodiment of the present application, the flowchart of the movement training recommendation method for the risk population of disability is shown in the figure;

[0038] Figure 2A schematic diagram of multi-modal data fusion in the first embodiment of the present application;

[0039] Figure 3 A flowchart of generating a sports training program based on a collaborative filtering model in the first embodiment of the present application;

[0040] Figure 4 A structural diagram of a recurrent neural network model in the first embodiment of the present application;

[0041] Figure 5 A structural diagram of a sports training recommendation system for a risk population in the second embodiment of the present application. DETAILED DESCRIPTION

[0042] The technical content of the present application will be described in detail below in combination with the drawings and specific embodiments.

[0043] The technical concept of the present application is to combine the content-based recommendation method with the collaborative filtering model, so as to integrate the feature learning of patients or items and the item recommendation process into a unified framework. First, the hidden features of patients or items are learned by using various deep learning models, and a unified optimization function is constructed by combining the collaborative filtering model to perform parameter training, and then the trained model is used to obtain the final hidden vectors of patients and items, and thus the personalized sports training recommendation for patients is realized.

[0044] First embodiment

[0045] As shown in Figure 1 The sports training recommendation method for a risk population provided by the first embodiment of the present application at least includes the following steps:

[0046] Step one: obtain the disease type and clinical characteristics, demographic information, sports research information and sports behavior characteristics of the patient to fuse and construct a role image and obtain static characteristics; wherein the patient is a risk population;

[0047] The basic model in the sports training recommendation for a risk population is used to construct a core image using the disease type and clinical characteristics of the patient, and a basic image is constructed using the demographic information, sports research information and sports behavior characteristics of the patient, and then the core image and the basic image are fused to construct the role image of the patient. In this process, static characteristics are obtained.

[0048] It should be noted that the construction method of the above-mentioned basic model can refer to the prior patent application with the application number 202411764001.3 and the name of "image construction method for a risk population, sports training recommendation method and system", which is briefly described as follows:

[0049] The basic model is constructed based on the following steps using data of a plurality of patients:

[0050] S1: Determine image mapping key elements according to pathological mechanisms;

[0051] S2: Construct a core image based on the key elements;

[0052] S3: Construct a basic image according to demographic information, sports research information and sports behavior characteristics;

[0053] S4: Construct a role image based on the core image set and the basic image set.

[0054] Among them, the key elements include cognitive cortex damage, physical function damage, and emotional control damage; the attributes of the core image of the disability disease include cognitive cortex damage, physical function damage, and emotional control damage, different attributes include different labels, and the label data is derived from the diagnosis information or research information of the patient, etc. The sports research information includes sports preferences, etc.

[0055] As shown in Figure 2 , cognitive cortex damage can cause cognitive impairment, which is manifested in complex attention, executive function, learning and memory, language, etc. Physical function damage can cause motor impairment, which is manifested in muscle strength, muscle tension, coordination ability, balance ability, etc. Emotional control damage can cause mental behavior disorders, which are manifested in hallucinations / delusions or depression / anxiety in the psychological and mental aspects, and are also manifested in aggression / hoarding or depression / wandering in the behavior aspect. Through the evaluation of the health status (disorders or diseases) of these patients, the labels of the patients can be mined.

[0056] The labels of the cognitive cortex damage of class A at least include:

[0057] A1. Frontal lobe damage: the patient can have manifestations of mental disorders such as memory decline and expression apathy, and can also have manifestations such as epilepsy, hemiplegia, motor aphasia, high fever, excessive sweating, abnormal vision and olfaction, etc.

[0058] A2. Parietal lobe damage: the patient can have manifestations of contralateral limb complex sensory disorders, body image disorders, and calculation disorders, etc.

[0059] A3. Temporal lobe damage: the patient can have manifestations of sensory aphasia, naming aphasia, olfactory hallucinations, epilepsy, mental abnormalities, memory disorders, visual field changes, etc.

[0060] A4. Occipital lobe damage: the patient can have visual impairment.

[0061] A5. Insular lobe damage: the patient can have manifestations of increased salivation, nausea, satiety, and other visceral motor and sensory disorders.

[0062] A6. Limbic leasion: Patients can have emotional and memory abnormalities, hallucinations and other mental abnormalities, and visceral activity disorders, etc.

[0063] A7…

[0064] The labels of the B class of physical function impairment at least include:

[0065] B1. Headache

[0066] B2. Nausea or vomiting

[0067] B3. Fatigue or drowsiness

[0068] B4. Tremor, stiffness and slowness of movement

[0069] B5. Dizziness or loss of balance

[0070] B6. Sensory problems, such as blurred vision, tinnitus, bad breath or changes in smell

[0071] B7. Sensitivity to light or sound

[0072] B8…

[0073] The labels of the C class of emotional control impairment at least include:

[0074] C1. Emotional changes or emotional fluctuations

[0075] C2. Depression or anxiety

[0076] C3. Difficulty falling asleep

[0077] C4. Drowsiness than usual

[0078] C5…

[0079] The attribute features of the core image are quantified and labeled to construct the core image. The attributes of the core image can be one-to-one corresponding to the core key elements, or more than the core key elements. Multiple labels of each attribute constitute a label library, and different label libraries represent different courses, impairment types and disability characteristics. In this step, the core image label of the patient is determined according to the type and clinical characteristics of the patient's diagnosed disease.

[0080] Demographic information, exercise research information, exercise behavior characteristics and social participation information are obtained through investigation or medical records, etc. to obtain the patient's label and construct the basic image.

[0081] Demographic information includes gender, age, education level, etc. Sports research information includes sports preferences and activity capabilities, etc. Sports preferences include sports types (e.g. skipping, running, etc.), sports venues (e.g. gyms, parks, etc.), sports atmosphere (quiet, dynamic, etc.). Activity capabilities include BADL (basic activity of daily living) and IADL (instrumental activity of daily living).

[0082] Sports behavior characteristics include sports frequency and sports cycle, etc.

[0083] Social participation information includes participation activities within the family and participation activities in society. Participation activities within the family include family decision-making, child-rearing, etc. Participation activities in society include work and study, entertainment activities, etc.

[0084] The basic portrait is not a core element affecting the occurrence of disability, but as a basic feature, it has a synergistic effect on the information of the core portrait of the patient's disease course, injury type, disability characteristics, etc. Therefore, it is included in the portrait system as a potential influencing variable for analysis.

[0085] The attributes of the basic portrait include at least:

[0086] Natural attributes: gender attribute is a widely used label. Different gender groups have obvious differences in preferences for different content. Moreover, through natural attribute labels such as age, region, education, occupation, marital status, and child status, it is easier to analyze the basic proportion of the patient group;

[0087] Vertical attributes: reflect the patient's sports needs, such as different types of sports preferences, etc.

[0088] Training attributes: training attributes are also an important attribute category, which helps to determine the patient's sports intention, sports cycle, and sports frequency.

[0089] Label attributes: when a patient starts using the system and generates the first piece of data, the system can assign the first label to it - new person. After that, with the accumulation of product patients, low-frequency patients, active patients, and high-frequency patients can be gradually distinguished.

[0090] Quantify and calibrate the basic attribute characteristics to form a basic attribute portrait set.

[0091] Through the above steps, a core portrait set and a basic portrait set are constructed for each patient. Next, based on the core portrait set and the basic portrait set, a role portrait needs to be constructed.

[0092] The core image and the attribute image are in a parallel relationship, the core image is the core feature of the individual, and the attribute image is the basic feature of the individual. The role image takes the basic attribute image as a one-dimensional structure and the core image as a two-dimensional structure. With the progress of time, the features of the core image at different time points (different disease development periods) form a three-dimensional structure. In this step, the image set is added with weight difference to form the role image of the individual.

[0093] The image sets constructed based on different features are merged according to predetermined rules. For example, different weights are given according to the importance of different features, and then the features in the two image sets are added with weights to merge the features. Here, the weight of the feature in the core image is greater than the weight of the feature in the basic image. Specifically, factors to be considered in weight distribution include feature coverage of each image set, consistency of data, rationality of weight distribution, etc., to ensure that the image after superposition can accurately reflect the comprehensive features of different patient groups and will not lose balance due to overemphasis on certain features.

[0094] The basic image set is taken as the coarse-grained feature of the individual, and the core image is taken as the fine-grained feature of the individual. By merging features of different granularities, the feature information of the user can be more comprehensively captured.

[0095] Coarse-grained features refer to relatively broad and general features, which provide general information about individuals or objects. These features include some basic attributes such as age, gender, occupation, etc., which are helpful for understanding the basic profile of individuals or patients. This feature is mainly applied to scenarios that quickly locate the overall profile of the individual, and is used as a basic data classification feature label. These features help with preliminary screening and classification, providing direction for subsequent in-depth analysis.

[0096] Fine-grained features refer to more specific and detailed features, which provide deeper information about individuals or objects. These features include some specific behaviors, interests, habits, etc., which are very important for understanding the uniqueness and personalized needs of individuals or objects. Fine-grained features are suitable for areas that require in-depth understanding of the uniqueness of individuals or objects, such as personalized recommendation systems, precise training, and behavior feedback analysis. These features can capture subtle differences and provide more accurate calculation and push support.

[0097] Therefore, coarse-grained features provide general information about users, while fine-grained features provide more specific details. This combination of coarse and fine features can improve the accuracy of image mapping; the fine-grained feature selection module can extract key local and fine-grained discriminative features from the object, while the coarse-grained feature selection module can obtain coarse-grained diversity features that provide context information, improving the discriminability of the model; by combining features of different granularities, the model can better generalize to different scenarios and conditions.

[0098] Specifically, the base image set (coarse-grained features) and the core image set (fine-grained features) are merged and superimposed. In this process, different weights need to be assigned to different features. When giving different weights according to the importance of different features, it is necessary to meet: 1) the feature weight in the core image is greater than the feature weight in the base image; 2) the feature coverage, data consistency, and rationality of weight allocation of each feature in the core image; 3) the feature coverage, data consistency, and rationality of weight allocation of each feature in the base image.

[0099] The core attribute features are quantified and calibrated one by one to form a core attribute image set.

[0100] The role image is a three-dimensional mapping system that integrates the features of the base image and the core image. The attributes of the core image include at least cognitive impairment, physical function impairment, and emotional control impairment. When obtaining the role image, it is necessary to construct a three-dimensional time slice based on the base image and the core image. This three-dimensional time slice is obtained by the following steps:

[0101] First, stage division. According to the progress of different types of senile disability diseases, the course of the disease is divided into several stages, including early, middle and late stages. Each major stage can be further subdivided into more specific sub-stages to better understand the development of the course of the disease.

[0102] Second, feature extraction. In each stage, the key features of the patient are extracted, including cognitive function, emotional state, and motor ability, etc. These features can be measured by standardized assessment tools, such as Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), and Berg Balance Scale, etc.

[0103] Finally, the construction of the time axis. The time axis from the discovery of the disease (for example, every month, every 2 months, every year, etc.) is used to represent the development of the course of the disease over time. On the time axis (as a time dimension), the starting point and the ending point of each stage can be marked, as well as the key features of each stage. The time axis can be linear or non-linear to reflect the complexity and uncertainty of the course of the disease. Such a time axis helps to better understand the progress of the course of the disease and provides guidance for treatment and intervention.

[0104] According to the industry consensus, the time sequence dimension characteristics in the role image are quantified, and image fusion is performed. The knowledge graph information such as the experience of clinicians, industry diagnostic standards, and expert consensus in the field is included, and the disease development stage is taken as the time dimension division node. The different disease development stages, quantified clinical characteristics, and other dimension information in the role image are associated. Taking senile dementia as an example, according to the industry consensus, the disease development can be divided into early stage (symptomless cerebral amyloid stage: SCD, MCI stage), middle stage (amyloid positive + synaptic dysfunction and (or) neurodegeneration stage), and late stage (amyloid positive + evidence of neurodegeneration + severe cognitive decline). The corresponding exercise training programs are different in different disease development stages. The content of the exercise training program is as shown in Figure 2

[0105] The different disease development stages, quantified clinical characteristics, and other dimension information in the role image are associated, which are used as the multi-dimensional image feature input of the pushed exercise training program, and preparation is made for the push of precise exercise training.

[0106] Step two: based on the role image, the first exercise training program is generated and pushed.

[0107] Based on the role image, the first exercise training program is automatically generated. The first exercise training program can be generated according to a pre-trained model (such as the collaborative filtering model mentioned later) or designed according to the experience of doctors.

[0108] Step three: collect structured data, text data, and image data during exercise training.

[0109] The structured data, text data, and image data of the patient exercising according to the first exercise training program are collected. According to the structured data, text data, and image data (imaging data or image data), the independent variables of the collaborative filtering model (deep learning model) are used as input. Specifically, including but not limited to:

[0110] ■Structured data: exercise diagnosis, life ability evaluation scale, demographic characteristic data, etc.

[0111] Text data: clinical diagnosis, case report data, etc. in the past two years (including the exercise diagnosis data of the patient obtained through the case, laboratory report, etc., such as upper limb motor function impairment);

[0112] ■Image data: magnetic resonance data, CT image data, etc.

[0113] ​Step four: input the data obtained in step three into the collaborative filtering model, combined with the role portrait, to generate and push the next exercise training plan.

[0114] In one aspect, multiple autoencoders and collaborative filtering networks are used in the collaborative filtering model, and graph embedding technology is used to map entities (such as patients, exercise types, physiological indicators, etc.) and relationships in the graph structure to a low-dimensional vector space. This step is achieved by deep learning embedding technologies such as Bayesian embedding model (TransR), Bayesian stacked denoising autoencoder (SDAE), Bayesian stacked convolutional autoencoder (SCAE), etc. Structured data, text data, and image data are converted into structured vectors, text vectors, and image vectors, respectively.

[0115] Specifically, the autoencoder is a layer-by-layer unsupervised learning model that mainly includes decoding and encoding processes, and is used to process high-dimensional data as a feature extraction tool to enable the collaborative filtering model to learn low-dimensional representations (hidden vectors) of users and items. These hidden vectors can capture the intrinsic relationship between users and items, thereby improving the performance of the recommendation system. Here, the autoencoder fuses three types of information (structured data, text data, and image data) to achieve multi-modal information fusion analysis and obtain item hidden vectors. Moreover, based on the role portrait data, the autoencoder uses feature extraction to obtain user hidden vectors.

[0116] On the other hand, in order to fully learn the features of auxiliary information, the collaborative filtering model respectively uses Bayesian stack denoising autoencoder (SDAE) to learn the vector representation of text information, Bayesian stack convolutional autoencoder (SCAE) to learn the vector representation of image information, and Bayesian embedding model (TransR) to learn the vector representation of structural information. SDAE is a deep learning model that extracts high-level features of data through multiple layers of unsupervised learning, which can capture semantic information and structural information in text, which is particularly useful for dealing with noise and uncertainty in text data. SCAE is used to learn the vector representation of image information because convolutional neural network (CNN) has an advantage in processing image data, which can capture the spatial hierarchy and local features of images. SCAE can learn more abstract and advanced image features by stacking multiple convolutional layers and pooling layers, which helps the model make more robust predictions when facing the diversity and complexity of image data. Learning structural information using TransR is because TransR is an embedding model designed specifically for knowledge graphs, which can handle complex relationships in knowledge graphs. TransR can effectively capture the multidimensional relationships between entities by mapping entities and relationships to the same vector space, and can handle incomplete information and noise in knowledge graphs. In addition, the flexibility and scalability of TransR make it suitable for different structured data, thereby improving the accuracy and robustness of the model in processing structured data. By converting text, images and structured data into unified vector representations, it can ensure that data from different sources has comparability in the feature space.

[0117] After feature extraction, graph embedding technology is beneficial to build individual data space, integrate multi-dimensional data such as time, movement preference, physiological and psychological characteristics and post-training state together, and form a four-dimensional data structure. Further, it integrates structure vector, text vector and image vector to form project hidden vector. Project hidden vector integrates features of multiple data types and can more comprehensively represent the characteristics of the project.

[0118] Specifically, as shown in Figure 3 the next training program is generated and pushed by using the collaborative filtering model, including the following sub-steps:

[0119] 1) Determine the key variables as one-dimensional data, two-dimensional data, three-dimensional data

[0120] First, it is necessary to determine which variables are important for assessing and predicting the recovery process of patients. By combining cognitive tasks (such as memory, attention tasks) and motor tasks (such as walking, balance training), the motor function and balance ability of patients can be improved. Through exercise, the blood flow and oxygen levels in the brain of patients are improved, activating neurons in a dormant state, promoting the growth and connection of neurons, forming specific nerve conduction patterns, and regulating neurotransmitter levels; while the improvement of cognitive function helps to improve the motor skills and efficiency of patients, enhance motivation and self-control (control ability of emotions, speech, etc.), and optimize motor strategies.

[0121] In summary, for patients with different diseases (different damaged brain regions), different key independent variables need to be determined.

[0122] For all patients with cognitive diseases, the following are important common key independent variables:

[0123] Time (T)

[0124] Type of Exercise

[0125] Intensity

[0126] Duration

[0127] Frequency

[0128] Patient Status

[0129] Physiological Indicators (such as heart rate, blood pressure, etc.)

[0130] Psychological Indicators (such as anxiety level, depression level, etc.)

[0131] Recovery Progress

[0132] For patients with different diseases, different key variables are determined, for example, including:

[0133] For patients with Alzheimer's disease:

[0134] Brain injury MRI ROI indicators (Region of Interest) hippocampus, medial temporal lobe, parietal lobe and frontal lobe

[0135] Cognitive Ability Scores Memory decline, spatial disorientation, attention deficit and executive function disorders, etc.

[0136] Parkinson's disease patients:

[0137] Brain injury MRI ROI indicators (Region of Interest) substantia nigra, striatum, prefrontal cortex, and parietal lobe

[0138] Activity of Daily Living Indicators (ADLIs) motor impairment, cognitive decline, and sensory abnormalities

[0139] Hormone Level Score (HLS) dopamine levels, etc.

[0140] 2) Create a data matrix based on key variables

[0141] Based on the key variables for each patient, create a data matrix where each row represents a data record at a time point and each column represents a variable. For example:

[0142] One-dimensional data: [Time (T)]

[0143] Two-dimensional data (exercise preference): [Type of Exercise

[0144] Intensity

[0145] Duration

[0146] Frequency]

[0147] Three-dimensional data (physiological and psychological characteristics): [Heart Rate

[0148] Blood Pressure

[0149] Anxiety Level

[0150] Depression Level]

[0151]

[0152] 3) Add lag features

[0153] To capture dependencies in time series, lag features can be added.

[0154] Four-dimensional: [Patient Status after Training Recovery Progress]

[0155] For example:

[0156] Time Patient status after training Rehabilitation progress T1 (e.g., 10:00 AM on October 3, 2024) Good 5％ T2 (e.g., 3:00 PM on October 3, 2024) Good 5％ T3 (e.g., 10:00 AM on October 4, 2024) Good 5％ T4 (e.g., 3:00 PM on October 4, 2024) Improved 10％ ... ... ...

[0157] Based on the aforementioned one-dimensional data, two-dimensional data, three-dimensional data, four-dimensional data, a four-dimensional data structure is formed. That is, the one-dimensional data is taken as the W axis; the two-dimensional data is taken as the Y axis, the three-dimensional data is taken as the X axis, and the four-dimensional data is taken as the Z axis, to construct an individual data space. The one-dimensional data includes a time variable, and the two-dimensional data includes multiple variables describing movement preferences; the three-dimensional data includes variables describing physical and psychological characteristics; and the four-dimensional data includes post-training state data.

[0158] 4) Convert to feature vector using embedding technology.

[0159] Use the hidden layer of the deep learning model for internal processing and fusion, and convert the four-dimensional data into a feature vector representation through embedding technology.

[0160] As shown in Figure 3 , by structured embedding, the structured data in the aforementioned individual data space of each patient is converted into a structure vector. For example, using the Bayesian TransR knowledge graph embedding method, entities and relationships can be converted into vectors, which can capture the complex relationships between entities.

[0161] By text embedding, text type data is converted into a text vector. For example, using a Bayesian sparse autoencoder, a low-dimensional representation of text data can be learned to capture the semantic information of the text.

[0162] By image embedding, image type data is converted into an image vector. For example, using a Bayesian sparse convolutional autoencoder, the features of an image can be extracted and converted into a vector.

[0163] Using the aforementioned Bayesian TransR knowledge graph embedding method, Bayesian sparse autoencoder, and Bayesian sparse convolutional autoencoder, the structure vector, text vector, and image vector are integrated into an item latent vector. This item latent vector is a vector that combines features of multiple data types and can more comprehensively represent the characteristics of different types of patients. It is a vector that can represent the characteristics of an item.

[0164] 5) Based on the feature vector, generate an item latent vector and a user latent vector for collaborative filtering learning to recommend a movement training program.

[0165] Use the attribute fusion graph convolutional network (AF-GCN) to fuse feature vectors from different modal data, and use the attention mechanism to weight the importance of different modal data, and gradually layer the feature vectors into latent vectors.

[0166] The embodiment of the present application adopts a collaborative filtering model combining graph structure and deep learning technology. The knowledge base is represented by using graph structure, which includes entities (such as sports categories, user attributes, user behaviors) and their relationships, so that the interaction data between users and items can be captured, and user preferences and item characteristics can be learned by using graph structure.

[0167] The multi-modal features (item latent vectors) and role image features (user latent vectors) extracted for the patient case are used as independent variables, and the first exercise training plan (for example, a sports prescription clinically prescribed by a doctor) in the previous step is used as a dependent variable, which is input into the collaborative filtering model to generate the next exercise training plan. The loss function of the collaborative filtering model is Cosine Contrastive Loss, which is used to maximize the similarity between positive sample pairs and minimize the similarity of negative sample pairs under margin constraints.

[0168] As described above, by using graph embedding technology, each entity and relationship in the graph structure of the individual data space is mapped to a low-dimensional vector space, which makes them can be calculated and compared, so that the collaborative filtering model can analyze the interaction data between users and items, capture the complex nonlinear relationship between patients and items, learn user preferences and item characteristics, and predict the items that users may be interested in.

[0169] It should be noted that the collaborative filtering model provided by the embodiment of the present application itself has a cold start problem. That is, for new patients, due to the lack of their historical behavior data, it is difficult to calculate the similarity between patients according to the conventional collaborative filtering model, and it is also difficult to provide personalized recommendations for them. This results in that when the collaborative filtering model is just used in the hospital, it is difficult for the model to provide accurate personalized recommendations for users. However, for structured data, image data, text data, etc., the collaborative filtering model can analyze the behaviors of other similar patients or similar items to predict the preferences of new users or new items, and only a small amount of exercise training plan data with clinically authoritative diagnosis is required to achieve the recommendation of exercise training plans for similar feature groups (cold start function), without the need for a large amount of training. This is because in the collaborative filtering model, based on the role image data and multi-modal data of new patients, similar patients and similar items can be easily found.

[0170] That is, the collaborative filtering model plays a role in the advantages of role profiling and multi-modal data fusion. For example, for patients with cognitive cortex damage, the magnetic resonance image will show abnormal signals in the specific brain area, and the clinical diagnosis report will also mention similar cognitive dysfunction. In the multi-modal data related to cognitive cortex damage and physical function damage, there is a high similarity in the characteristics. Through the collaborative filtering model, the commonness and regularity of different patients in key features can be mined. This model can effectively utilize the similarity features in multi-modal data, thereby improving the accuracy of feature similarity judgment. For example, although two patients are different in emotional control damage, if they are highly similar in cognitive cortex damage and physical function damage, then in the exercise training recommendation, these two patients are more likely to be classified as similar patients, and then suitable exercise programs are recommended for new patients. Using the core profile with fine granularity, the collaborative filtering model can accurately grasp the key needs of new patients and find similar patients and projects.

[0171] According to the rich information in the role profile, match and filter from multiple aspects to find a patient group similar to the new patient in multiple dimensions, and then recommend a more suitable exercise training program for the new patient, improving the individualization of the recommendation. For example, for patients with severe cognitive cortex damage, more emphasis is placed on cognitive training and brain function activation exercise programs; while patients with prominent physical function damage focus on training for the recovery and enhancement of limb motor function. This personalized recommendation method can better solve the cold start problem than the traditional collaborative filtering model that only relies on historical behavior data, providing accurate and effective exercise training recommendations for new patients and helping them improve their cognitive, emotional, motor, and speech abilities.

[0172] In addition, based on the basic profile such as medical history, demographic characteristics, and the core profile such as examination results and doctor's recommendations, the generated patient state sequence data, and the fusion process of different modal data such as imaging data (image data) and scale data (text data), the use of the collaborative filtering model can better grasp the association between different media modalities through multi-layer attribute fusion, improving the accuracy and generalization ability of exercise program recommendation.

[0173] Step five: collect structured data, text data, image data, and clinical diagnosis data during exercise training multiple times in the manner of step three to obtain exercise training cycle data.

[0174] The patient exercises multiple times according to the exercise training program, and accumulatively collects data (structured data, text data, image data) in multiple exercises in the manner of step three to obtain exercise training cycle data.

[0175] The training cycle data in the training program is generally divided into multi-year cycle data, large cycle data, medium cycle data and small cycle data.

[0176] ■Multi-year cycle: Due to the large individual differences among the elderly, specific problems need to be analyzed. For example, (1) The physical condition and recovery ability of each elderly person is different, so the exercise rehabilitation cycle will vary from person to person. Generally speaking, the elderly with better physical condition and lighter disability degree may recover faster, and can take 2 years as a cycle. (2) The more serious the disability degree, the longer the recovery time usually takes. Elderly people who are completely bedridden take 8 years as a cycle, and elderly people who can partially take care of themselves take 4 years as a cycle. During the rehabilitation process, the physical condition and rehabilitation progress of the elderly need to be continuously monitored and adjusted as needed. This helps to ensure the effectiveness of the rehabilitation plan and shorten the rehabilitation cycle as much as possible.

[0177] ■Large cycle generally takes this year or one year as the time limit to develop a half-year or annual training plan. There are also three large cycles in a year, which is generally related to the classification of sports projects.

[0178] ■Medium cycle is called stage and monthly training in training practice, generally 4-8 weeks, to develop a stage or monthly training plan.

[0179] ■Small cycle is called weekly training in training practice, usually with a calendar week as the deadline to develop a weekly training plan. It can also be arranged for 4-10 days. For example, if a small recovery adjustment can be completed in 4 days, a 4-day recovery small cycle can be arranged.

[0180] Step six: determine whether the exercise training program is the Nth, if so, go to step seven; if not, return to step four, and generate the next exercise training program according to the characteristics in step five.

[0181] N is a preset number of times, for example, 3 times. Through this step, the dynamic characteristics of the patient when training according to multiple exercise training programs can be obtained, and the characteristics have a time sequence dimension. Such design is conducive to the recurrent neural network model (Recurrent Residual Network) to learn more complex dynamic evolution representation.

[0182] Step seven: based on the data collected by the patient in step three and step five, input the recurrent neural network model together to iteratively optimize the exercise training program.

[0183] In this step, models such as Figure 4The recurrent neural network model shown can better learn more complex dynamic evolution representations. In the recurrent neural network model, the historical interaction information between the patient and the exercise is the key data driving the change of the patient's preference and the exercise state, so using the co-evolution model can capture the evolution hidden representation of the patient and the exercise. Here, the recurrent neural network model uses the training cycle vector input into the recurrent neural network model to learn the dynamic feature representation of the patient and the exercise using the recurrent neural network, and obtains a dynamic evolution representation with a time sequence relationship. The training cycle vector refers to converting various attributes (such as time, frequency, intensity, etc.) in the training cycle data into numerical features, and then converting the obtained vector through embedding or encoding technology.

[0184] As shown in Figure 4 , the recurrent neural network model uses two long short-term memory networks (Long short-term memory, abbreviated as LSTM) based on the traditional recurrent neural network (RNN) to learn the time sequence change of the patient's exercise preference and the long-term (for example, seasonal) evolution of the exercise, respectively. In the figure, y i , which represents the dynamic feature vector of patient i at the time sequence (time t, t+1, etc.); y j , which represents the dynamic feature vector of exercise j at the time sequence. Moreover, y i and y j are both affected by the static feature u i of the patient and the static feature m j of the exercise. This is to consider the long-term exercise preference of the patient and the static attribute of the exercise, so that the recurrent neural network model learns the static hidden representation of the patient and the static hidden representation of the exercise at the same time.

[0185] In other words, at each time step, the recurrent neural network model uses one LSTM network to combine the dynamic feature vector of the previous time step of the patient and the current static feature vector to update the dynamic feature vector; and uses another LSTM network to combine the dynamic feature vector of the previous time step of the exercise and the current static feature vector to update the dynamic feature vector. The recurrent neural network model that combines static features and dynamic features can simultaneously consider the long-term exercise preference of the patient and the short-term change, and the long-term characteristics and short-term change of the exercise training program.

[0186] Specifically, the recurrent neural network model uses an LSTM-based recurrent neural network to model the dynamic changes of the patient and the exercise training program. For patient i and exercise j (an exercise training program includes multiple exercises), it is assumed that u i represents the static feature vector of patient i, and m j represents the static feature vector of exercise j. At time t, it is assumed that uit is the dynamic feature vector of patient i (i.e. Figure 4 i，t ), m jt is the dynamic feature vector of exercise j (i.e. Figure 4 j，t ). The dynamic features u i，t+1 and m j，t+1 at time t+1 can be solved by a LSTM network sequentially, as follows:

[0187] u i,t+1 = g(u it , {r ij|t})

[0188] m j，t+1 = h(m jt , {r ij|t})

[0189] where g and h are functions to be learned, g represents the function in the LSTM network for updating the dynamic feature of the patient; h represents the function in the LSTM network for updating the dynamic feature of the exercise training program; r ij|t represents the score of patient i at time t on exercise j. The dynamic feature u i,t+1 of the patient at time t+1 is calculated by the LSTM network according to the dynamic feature u i,t of the patient at time t and the response r ij|t . The dynamic feature m j,t+1 of the exercise at time t+1 is calculated by the LSTM network according to the dynamic feature m j，t of the exercise at time t and the response r ij|t . This shows that the dynamic features of the patient and the exercise training program are updated by the LSTM network according to their state at the previous time point and the current response. Therefore, the recurrent neural network model can capture the dynamic features that change over time, thereby better simulating and predicting the patient's response to different exercise training programs.

[0190] The dynamic feature and static feature of patient i (u i，t+1 , m j，t+1 , u it , m jt ) are input into the recurrent neural network model, which can predict the current interest (score) of the patient, r ij The predicted value of r where f is also a function to be learned.

[0191] The loss function of the recurrent neural network model is the sum of the square error loss function and the regularization term. Adjusting the parameters θ to minimize the loss function can achieve the overall optimization of the model. That is, the following function is used for optimization:​​

[0192]

[0193] It can be seen that the optimization objective of the recurrent neural network model is to make the pushed exercise training scheme more and more consistent with the current exercise preference, exercise level and exercise state of the patient, that is, the parameters generated through training make the prediction close to the actual value. θ represents the parameters to be learned, is a set of (patient, exercise training scheme, time sequence) tuples observed in the training set, and R represents a regularization function.

[0194] Although the objective function and building block in the recurrent neural network model are very standard, simple back propagation application cannot easily solve this optimization problem. The key challenge is that the score of each patient depends on the patient state and exercise training scheme. Back propagation through two sequences is computationally prohibitive. This problem is alleviated by back propagating the gradient from the patient's exercise feedback, but each score still depends on the patient state, which in turn acts on the complete sequence of exercise training scheme pushing.

[0195] In an embodiment of the present application, the recurrent neural network model adopts an alternating subspace descent strategy, in which the gradient is not propagated into the exercise training scheme sequence assuming that the exercise state is fixed, and the parameters of the patient sequence are updated by back propagating the gradient of all scores of the patient; subsequently, the patient sequence is updated and the exercise training scheme sequence is updated alternately. In this way, only one standard forward propagation and back propagation can be performed for each patient's exercise training scheme pushing, and finally the optimization of the recurrent neural network model is realized.

[0196] As can be seen from the above, the present application uses a hybrid recommendation model (collaborative filtering model and recurrent neural network model) as the core tool for exercise training recommendation. Compared with the traditional content recommendation model, the collaborative filtering model in the hybrid recommendation model can simultaneously learn the representation of multi-source data (structured data, text data, image data, training cycle data) and utilize multi-dimensional information such as structure information, content information and image information. At the same time, the recurrent neural network in the hybrid recommendation model is used to capture the long-term preference evolution representation and short-term preference representation of the patient. Therefore, by fusing different models, the exercise dose, exercise state and exercise mode of a specific patient can be effectively evaluated, and finally the best exercise training scheme can be pushed.

[0197] At each training of the patient, the data collected during the previous exercise training is used to output a new exercise training scheme in the manner of step seven. This is repeated until a predetermined goal is reached, such as recovery or maintenance of cognitive, emotional, exercise, speech and other abilities.

[0198] As an optional step, after step seven, step eight can also be added: according to the data in the last exercise training program, evaluate the training effect, if the effect is improved, make and output the exercise training report or directly end; if the effect is not improved, return to step seven to regenerate the exercise training program. The method of evaluating the training effect can be evaluated according to the conventional method, which can be evaluated by a doctor, or the data in the last exercise training program can be compared and evaluated with the data in the previous exercise training program.

[0199] Compared with the prior art, the present application has the following technical advantages by formulating an individualized targeted exercise training program:

[0200] 1) A exercise recommendation framework combining the advantages of multiple model algorithms is proposed. On the one hand, the framework combines collaborative filtering and auto-encoding learning technology to build a fusion analysis system for multi-modal data, which can effectively integrate multi-source information into the recommendation model, thereby enriching the analysis source and providing more comprehensive data support for exercise recommendation. On the other hand, the recurrent neural network is introduced to effectively calculate the time series data. Through the retrospective analysis of the patient's exercise data at different time periods, the coherence and sustainability of the exercise recommendation are ensured, and the patient's compliance is improved.

[0201] 2) A training cycle data system is constructed, which divides the exercise training process into annual cycles, large cycles, medium cycles and small cycles, and generates corresponding time series data sets for different cycles. When these data are introduced into the model for exercise recommendation calculation, the time series dimension feature is increased. This enables the model to more fully and regularly utilize historical exercise data, thereby assisting in generating periodic exercise recommendation programs.

[0202] 3) A deep learning-based exercise pushing system is constructed. The system integrates multi-dimensional exercise training pushing logic into a complete process, which can continuously update exercise pushing content based on the patient's own multi-dimensional exercise data. In this way, it ensures that the training program always matches the patient's current state, thereby improving the training effect.

[0203] 4) A multi-level patient portrait system is constructed. The system gradually and deeply deconstructs and evaluates the patient's attribute features, behavior features, and comprehensive features (including cognition, emotion, exercise, speech, etc.), ensuring that the individual tag system has a detailed granularity. This provides a large amount of effective sequence information for subsequent input of patient data into the model for precise exercise recommendation, further improving the accuracy and individualization of the recommendation.

[0204] Second embodiment

[0205] On the basis of the above-mentioned exercise training recommendation method for the disabled risk population, the second embodiment of the present application further provides an exercise training recommendation system for the disabled risk population. As shown in the accompanying drawings, the exercise training recommendation system comprises one or more processors and memories. Figure 5 The memory is coupled with the processor, and is configured to store one or more programs. When the programs are executed by the processor, the processor implements the exercise training recommendation method for the disabled risk population in the above-mentioned embodiments.

[0206] The processor is configured to control the overall operation of the exercise training recommendation system to complete all or part of the steps of the above-mentioned exercise training recommendation method for the disabled risk population. The processor can be a central processing unit (CPU), a graphics processing unit (GPU), a field programmable logic gate array (FPGA), an application specific integrated circuit (ASIC), a digital signal processing (DSP) chip, etc. The memory is configured to store various types of data to support the operation of the exercise training recommendation system. For example, the data can include instructions for operating any application or method on the exercise training recommendation system, and application-related data. The memory can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, etc.

[0207] In another exemplary embodiment, the present application also provides a computer-readable storage medium comprising program instructions, which are executed by a processor to implement the steps of the exercise training recommendation method in any one of the above-mentioned embodiments. For example, the computer-readable storage medium can be the above-mentioned memory comprising program instructions, which can be executed by the processor of the system to complete the above-mentioned exercise training recommendation method for the disabled risk population and achieve the technical effects consistent with the above-mentioned method.

[0208] It should be noted that the above-mentioned embodiments are only illustrative. The technical solutions of each embodiment can be combined, and the order of each step can be changed, all of which are within the protection scope of the present patent.

[0209] The exercise training recommendation method and system for the disabled risk population provided by the present application are described in detail above. Any obvious modifications made by those skilled in the art without departing from the essential content of the present application shall constitute an infringement of the patent right of the present application and shall bear the corresponding legal responsibility.

Claims

1. A method for recommending exercise training for a person at risk of disability, characterized by It comprises the following steps: Step one: obtain the disease type and clinical characteristics of the patient, demographic information, exercise research information and exercise behavior characteristics to fuse and construct a role image and obtain static characteristics; wherein the patient is a disability risk population; Step two: generate and push the first exercise training program based on the role image; Step three: collect structured data, text data and image data during exercise training; Step four: input the data obtained in step three into the collaborative filtering model to generate and push the next exercise training program in combination with the role image; Step five: collect structured data, text data and image data during exercise training in the manner of step three to obtain exercise training cycle data; Step six: determine whether the exercise training program is the Nth, if so, go to step seven; if not, return to step four to generate the next exercise training program based on the characteristics in step five, N is a positive integer; Step seven: based on the data collected by the patient in steps three and five, input the recurrent neural network model to iteratively optimize the exercise training program, wherein the recurrent neural network model introduces two LSTM networks based on the recurrent neural network; at each time step, one of the LSTM networks combines the dynamic feature vector of the patient at the previous time step and the static feature vector at the current time step to update the dynamic feature vector of the patient; the other LSTM network combines the dynamic feature vector of the exercise at the previous time step and the static feature vector at the current time step to update the dynamic feature vector of the exercise, Wherein, the step four comprises the following sub-steps: 1) determine the key variables as one-dimensional data, two-dimensional data and three-dimensional data; 2) create a data matrix according to the key variables; 3) add lag features as four-dimensional data; 4) convert to feature vectors using embedding technology; 5) generate item hidden vectors and user hidden vectors based on the feature vectors for collaborative filtering learning to recommend exercise training programs, Use the Bayesian TransR knowledge graph embedding method to convert entities and relationships into vectors; use the Bayesian sparse autoencoder embedding method to convert text data into text vectors; use the Bayesian sparse convolutional autoencoder to extract image features and convert them into image vectors, so as to realize multi-modal data fusion based on the structured data, the text data and the image data, obtain the user hidden vector and the item hidden vector, and input them into the attribute fusion convolutional layer to recommend the next exercise training program.

2. The exercise training recommendation method for a disability risk-oriented population according to claim 1, characterized in that The recurrent neural network model is optimized using the following function: wherein, denotes the parameters to be learned; is a set of tuples observed in the training set, the tuples comprising a patient, an exercise training regimen, and a time series; denotes a patient at a time a score for an exercise ; is a prediction value for ; denotes a regularization function.

3. The exercise training recommendation method for disability risk population according to claim 2, characterized in that: The recurrent neural network model adopts an alternating subspace descent strategy, wherein it is assumed that the exercise state is fixed and the gradient is not propagated to the exercise training program sequence, and the parameters of the patient sequence are updated by backpropagating the gradient of all patient scores; subsequently, the patient sequence is updated and the exercise training program sequence is updated alternately.

4. The exercise training recommendation method for a disability risk-oriented population according to claim 1, characterized by The multi-modal features and role image features extracted according to the patient condition are taken as independent variables, and the exercise prescription suggestions clinically prescribed by the doctor are taken as dependent variables, which are input into the collaborative filtering model; the loss function of the collaborative filtering model is a cosine contrast loss, which is used to maximize the similarity between positive sample pairs and minimize the similarity of negative sample pairs under the margin constraint.

5. The exercise training recommendation method for a disability risk-oriented population according to claim 1, characterized in that Further comprising the following steps: According to the data in the last exercise training scheme, the training effect is evaluated, if the effect is improved, an exercise training report is made and output or directly ended; if the effect is not improved, the exercise training scheme is regenerated again in step seven.

6. A motion training recommendation system for a risk of disability population, characterized by The method comprises one or more processors and a memory, wherein the memory is coupled with the processor, and is used for storing one or more programs, when the programs are executed by the processor, the processor implements the exercise training recommendation method for the risk population as claimed in any one of claims 1-5.

Images