Intelligent rehabilitation device

Abstract

Embodiments of the present application disclose an intelligent rehabilitation device, aiming at solving the problems in the prior art that the position and height cannot be adjusted, the assessment accuracy is poor, the rehabilitation management efficiency is low and the like. The device comprises a mobile cart, and an embedded superordinate computer, a mobile unit and a wearable unit which are respectively mounted on the mobile cart. The mobile unit and the wearable unit are both connected to the embedded superordinate computer. The embedded superordinate computer is provided with a rehabilitation training module, a rehabilitation assessment module, a data acquisition module, a data processing module, a mobile control module and a storage module. The mobile unit comprises rollers, an electric push rod and a driver which are arranged on the mobile cart. The driver is connected to the mobile control module. The wearable unit comprises rehabilitation training gloves and a somatosensory instrument. The rehabilitation training gloves and the somatosensory instrument both perform rehabilitation training by means of the rehabilitation training module, and send rehabilitation training data to the data acquisition module.

Description

An intelligent rehabilitation device Technical Field

[0001] The embodiments of the present application relate to the field of medical technology, and specifically to an intelligent rehabilitation device. Background Art

[0002] With the improvement of living standards and the aging of the population, cardiovascular and cerebrovascular diseases have become a major health threat, and stroke is the leading cause of death and lasting disability. Clinical research and practice have confirmed that with timely and effective rehabilitation training, approximately 90% of patients can regain some level of daily movement and self-care abilities. Clinically, the 0-6 month recovery period is considered the golden recovery period, but rehabilitation is a long-term process, most lasting more than a year. Furthermore, the number of stroke patients is extremely large, and rehabilitation medical resources are extremely scarce, leading to a huge demand for intelligent management devices.

[0003] Current rehabilitation treatment usually involves hands-on assistance by rehabilitation therapists or one-on-one treatment with mechanical assistive devices. This method has many shortcomings: (1) low treatment efficiency and high labor costs; (2) rehabilitation assessment is highly subjective, resulting in an inability to guarantee the accuracy of rehabilitation training assessment results; (3) the selection of training movements is not close to life, making it difficult for patients to return to normal social life; (4) the position and height of the device are fixed, which is inconvenient for patients who cannot move around easily.

[0004] Summary of the Invention

[0005] The embodiments of the present application provide an intelligent rehabilitation device that can improve the efficiency of rehabilitation treatment for patients and at the same time improve the intelligence of the rehabilitation management device.

[0006] The technical solutions provided by the embodiments of this application are as follows:

[0007] An intelligent rehabilitation device, comprising: a mobile cart;

[0008] The mobile cart includes an embedded host computer, a mobile unit and a wearable unit, and the mobile unit and the wearable unit are both communicatively connected to the embedded host computer;

[0009] The embedded host computer includes a rehabilitation training module, a rehabilitation assessment module, a data acquisition module, a data processing module, a movement control module and a storage module;

[0010] The rehabilitation training module is used to provide a scenario training game designed based on rehabilitation movements, and the scenario training game is used to implement rehabilitation training for the patient's upper limbs, lower limbs and hands;

[0011] The rehabilitation assessment module is used to assess the motor functions of the upper limbs, lower limbs and hands of the rehabilitation training patient;

[0012] The data acquisition module is used to collect information about the rehabilitation training patients, the information including feature vectors corresponding to multiple dimensions of the patients, each feature vector is used to indicate the patient information under each dimension, the dimensions including: patient medical history, living habits, environmental factors and rehabilitation training status;

[0013] The data processing module is configured to predict the degree of influence of the patient information in each dimension on the patient's rehabilitation based on the feature vectors of each dimension, obtain a first prediction vector corresponding to the feature vector of each dimension, the first prediction vector being used to indicate the degree of influence of the patient information in each dimension on the patient's rehabilitation, and perform feature concatenation on the first prediction vectors to obtain a target vector; predict the rehabilitation effect corresponding to the patient information in each dimension based on the target vector to obtain a second prediction vector, the second prediction vector being used to indicate the rehabilitation effect corresponding to the patient information in each dimension; and generate prescription information for the patient's rehabilitation based on the second prediction vector, the prescription information including medication information and drug dosage;

[0014] The mobile control module controls the mobile cart to move and adjust its height by acquiring the position of the rehabilitation patient and measuring the distance between the mobile cart and the rehabilitation patient;

[0015] The mobile unit includes a roller, an electric push rod and a driver provided on the mobile cart, wherein the driver is connected to the mobile control module, and the mobile control module drives the roller to move and adjust the position and controls the electric push rod to adjust the height through the driver;

[0016] The wearable unit includes rehabilitation training gloves and a somatosensory instrument. Both the rehabilitation training gloves and the somatosensory instrument perform rehabilitation training through the rehabilitation training module, and send rehabilitation training data to the data acquisition module.

[0017] Furthermore, the data acquisition module includes a receiver and a camera, and the receiver is wirelessly connected to the rehabilitation training glove and the somatosensory instrument respectively;

[0018] The receiver is used to receive the motion data of the rehabilitation training gloves and the somatosensory instrument, and the camera collects the motion data by capturing the movements of the rehabilitation training patient;

[0019] The data processing module is further configured to process the motion data based on multi-source information fusion.

[0020] Furthermore, a human-computer interaction column is provided with a card reading area and a button area, wherein the card reading area is used to identify the rehabilitation training patient, and the button area is provided with a plurality of buttons, wherein the buttons are used to assist in the scenario training game;

[0021] A cloud server storing an exercise prescription;

[0022] The embedded host computer is used to download the exercise prescription or upload the training data through the cloud server;

[0023] The power supply module includes a magnetic ring, a transformer, a filter, and a data line, and is used to provide power and process power signals.

[0024] Furthermore, the rehabilitation training glove includes a glove body, a first shell installed on the glove body, a first battery, a first circuit board and a bending sensor are arranged in the first shell, and a first indicator light is arranged on the first shell.

[0025] Furthermore, the somatosensory instrument includes a strap and a second shell mounted on the strap, a second circuit board and a second battery are arranged in the second shell, and a second indicator light is arranged on the second shell.

[0026] Furthermore, the rehabilitation assessment module includes an assessment auxiliary unit and an assessment unit, and the assessment unit includes an intelligent scale assessment subunit and a compensatory movement quantitative assessment subunit;

[0027] The evaluation auxiliary unit is used to complete the task of recognizing rehabilitation training movements and obtain recognition results;

[0028] The evaluation unit is configured to obtain a corresponding evaluation result based on the recognition result.

[0029] Furthermore, the embedded host computer realizes terminal interconnection through a network for remote medical diagnosis or online rehabilitation training guidance.

[0030] Furthermore, the mobile control module is connected to a radar detector and a positioning module, the radar detector is used to detect the position of the rehabilitation training patient, and the positioning module is used to determine the current position of the mobile cart.

[0031] Furthermore, the mobile cart includes a base and a connecting block, the roller is arranged at the bottom of the base, the electric push rod is arranged in the connecting block, and the embedded host computer is connected to the top of the electric push rod.

[0032] Furthermore, the mobile control module is connected to an infrared sensor and an infrared signal transmitter. When the device is in use, position and height adjustment are achieved through the infrared sensor, the infrared signal transmitter and the mobile control module.

[0033] Compared with the prior art, the beneficial effects of the embodiments of the present application are:

[0034] 1. This embodiment of the present application provides an intelligent rehabilitation device comprising a mobile cart, an embedded host computer, a mobile unit, and a wearable unit, each mounted on the mobile cart. The embedded host computer controls the mobile unit to achieve intelligent height recognition and height adjustment.

[0035] 2. This embodiment of the present application provides an intelligent rehabilitation device that uses a rehabilitation training module in an embedded host computer to perform rehabilitation training tasks in the form of a virtual game. A rehabilitation assessment module provides visualized, quantitative rehabilitation assessment results. The device is simple to operate, highly efficient, and highly entertaining.

[0036] 3. This embodiment of the present application proposes an intelligent rehabilitation device that, through rehabilitation training gloves and a somatosensory instrument, can perform multi-faceted rehabilitation training for the upper and lower limbs, as well as the fingers. By identifying rehabilitation training movements and establishing a scale assessment and compensation assessment, this device provides a precise assessment of rehabilitation training movements. Compared to existing assessment systems and devices, the assessment results are more accurate. Furthermore, the assessment model can be learned, which has a broad application market.

[0037] 4. This embodiment of the application proposes an intelligent rehabilitation device that uses a human-computer interaction column to establish rehabilitation files, allowing users to monitor the rehabilitation progress of all patients at any time. This not only improves the user experience but also reduces the burden on staff through detailed file management, making the entire rehabilitation training process more standardized.

[0038] 5. An embodiment of the present application proposes an intelligent rehabilitation device, which loads a variety of rehabilitation training games through a rehabilitation training module, and selects interesting mini-games close to life to assist patients in rehabilitation training. The rehabilitation training module can fully train the patient's upper limbs, lower limbs and fingers, thereby accelerating the patient's rehabilitation progress.

[0039] 6. An embodiment of the present application proposes an intelligent rehabilitation device, which can predict the patient's rehabilitation effect by obtaining the patient's information in various dimensions and generate prescription information for the patient's rehabilitation based on the patient's information in various dimensions, thereby improving the patient's rehabilitation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] FIG1 is a structural diagram of an intelligent rehabilitation device according to an embodiment of the present application;

[0041] FIG2 is a second structural diagram of an intelligent rehabilitation device according to an embodiment of the present application;

[0042] FIG3 is a third structural diagram of an intelligent rehabilitation device according to an embodiment of the present application;

[0043] FIG4 is a schematic diagram of an intelligent rehabilitation device according to an embodiment of the present application;

[0044] FIG5 is a structural diagram of a rehabilitation training glove in an intelligent rehabilitation device according to an embodiment of the present application;

[0045] FIG6 is a structural diagram of a somatosensory instrument in an intelligent rehabilitation device according to an embodiment of the present application;

[0046] FIG7 is a fourth structural diagram of an intelligent rehabilitation device according to an embodiment of the present application.

[0047] The figure numbers are as follows: 1-mobile cart, 101-base, 102-connecting block, 11-embedded host computer, 111-data acquisition module, 1111-receiver, 1112-camera, 112-data processing module, 113-mobility control module, 114-rehabilitation training module, 115-rehabilitation assessment module, 116-storage module, 12-mobile unit, 121-driver, 122-electric push rod, 123-roller, 2-wearable unit, 201-rehabilitation training gloves, 202-somatosensory instrument, 13-human-computer interaction column, 14-power module. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the embodiments described below are part of the embodiments of the present application, not all of the embodiments. Generally, the components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0049] Therefore, the detailed description of the embodiments of the present application provided below in conjunction with the accompanying drawings is intended to represent only selected embodiments of the present application and is not intended to limit the scope of protection claimed in the present application. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present application without making any creative efforts shall fall within the scope of protection claimed in the present application.

[0050] It should be understood that in the description of the embodiments of the present application, the terms "first," "second," etc. are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, features defined as "first," "second," etc. may explicitly or implicitly include one or more of the described features.

[0051] In the description of the implementation methods of the embodiments of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection, or mutual communication; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the implementation methods of the embodiments of the present application can be understood according to specific circumstances.

[0052] FIG1 is a block diagram of the overall structure of the intelligent rehabilitation device provided in an embodiment of the present application. Referring to FIG1 , an embodiment of the present application provides an intelligent rehabilitation device, comprising:

[0053] The mobile cart 1 includes an embedded host computer 11 and a mobile unit 12 respectively installed on the mobile cart 1 , and a wearable unit 2 . Both the mobile unit 12 and the wearable unit 2 are connected to the embedded host computer 11 .

[0054] The embedded host computer 11 includes a processor and a memory. The processor is used to execute the programs of the data acquisition module 111, the data processing module 112, the movement control module 113, the rehabilitation training module 114, and the rehabilitation assessment module 115.

[0055] The processor may be a general-purpose processor (central processing unit, CPU), a microprocessor, an application-specific integrated circuit (application-specific integrated circuit, ASIC), or one or more integrated circuits for executing the program of the solution of the present application.

[0056] The memory may be the storage module 116 in the present application, and the memory may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0057] As shown in FIG2 , the wearable unit 2 includes a rehabilitation training glove 201 and a somatosensory instrument 202. Both the rehabilitation training glove 201 and the somatosensory instrument 202 perform rehabilitation training via the rehabilitation training module 114 and transmit the rehabilitation training data to the data acquisition module 111. The wearable unit 2 wirelessly communicates with the embedded host computer 11 via a ZigBee, Bluetooth, or Wi-Fi wireless communication protocol, enabling the wearable unit 2 to perform rehabilitation training according to the training content of the rehabilitation training module 114 and transmit the rehabilitation training data to the data acquisition module 111.

[0058] The data acquisition module 111 is used to collect information about the rehabilitation training patients. The information includes feature vectors corresponding to multiple dimensions of the patients. Each feature vector is used to indicate the patient information under each dimension. The dimensions include: patient medical history files, living habits, environmental factors and rehabilitation training status.

[0059] Among them, the patient's medical history file includes: patient's chief complaint, current medical history, past history, personal life history, marital history, family history, diagnosis records, treatment recommendations, surgical records, pathology examination reports, etc. Lifestyle habits include: diet, exercise, smoking history, drinking, work and rest, etc. Environmental factors include: temperature, humidity, air dust, wind speed, ambient noise decibels, air quality, etc. Rehabilitation exercises include: motor function, balance function, and cognitive function. Of course, in addition to these dimensions of factors that will affect the patient's recovery, there are other factors, such as: patient gender, place of residence, age, education level, family conditions, marital history, economic status, occupation, type of work, and work environment. Taking living habits as an example, including diet (normal: 0, salty: 1, bland: 2, oily: 3... etc.), whether to exercise (moderate exercise: 0, never exercise: 1, small amount of exercise: 2, large amount of exercise: 3, excessive exercise: 4...), whether to smoke, whether to drink, and whether the work and rest schedule is regular. The characteristic vector of the living habit dimension is (0, 0, 1, 1, 0, 0...).

[0060] The data processing module 112 includes a first prediction unit, a processing unit, a second prediction unit and a generation unit; the first prediction unit is used to predict the degree of influence of the patient information of each dimension on the patient's rehabilitation based on the feature vector of each dimension, and obtain a first prediction vector corresponding to the feature vector of each dimension, and the first prediction vector is used to indicate the degree of influence of the patient information of each dimension on the patient's rehabilitation; the processing unit is used to perform feature splicing on each of the first prediction vectors to obtain a target vector; the second prediction unit is used to predict the rehabilitation effect corresponding to the patient information of each dimension based on the target vector, and obtain a second prediction vector, and the second prediction vector is used to indicate the rehabilitation effect corresponding to the patient information of each dimension; the generation unit is used to generate prescription information for the patient's rehabilitation based on the second prediction vector, and the prescription information includes medication information and drug dosage.

[0061] Among them, the prescription information also includes the name of the medical institution, patient name, gender, age, outpatient or inpatient medical record number, department or ward and bed number, clinical diagnosis, date of issuance and other information; the name, dosage form, specifications, quantity, usage and dosage of the drug and other information.

[0062] Optionally, the first prediction unit is specifically used to: input the feature vector corresponding to each dimension into the first prediction model corresponding to each dimension, to obtain a first prediction vector corresponding to each feature vector, the first prediction vector being used to indicate the influence of the patient information of each dimension on the patient's recovery; wherein, each of the first prediction models is an LSTM-based neural network model, and each of the LSTM-based neural network models is a model trained using multiple patient information samples and corresponding label information.

[0063] The second prediction unit is specifically used to: input the target vector into a preset second prediction model to obtain a second prediction vector corresponding to each first prediction vector, and the second prediction vector is used to indicate the predicted result of the rehabilitation effect corresponding to the patient information of each dimension; wherein, the second prediction model is a neural network model based on the attention mechanism, and the neural network model based on the attention mechanism is a model trained using multiple samples of information on the influence of patient information of each dimension on patient rehabilitation and corresponding label information.

[0064] The processing unit is further configured to: obtain an activation function, the activation function including weight coefficients of each dimension and a preset bias constant; input the second prediction vector into the activation function, and determine at least one target subvector from the second prediction vector using the activation function;

[0065] The generating unit is further configured to generate prescription information for the patient's rehabilitation based on the at least one target sub-vector.

[0066] The processing unit is also used to: obtain a rehabilitation prescription map database, which includes multiple prescription information; generate corresponding prescription vectors for each prescription information in the rehabilitation prescription map database; obtain a matching function and a weight vector, wherein the matching function is used to match the corresponding prescription vector according to the target sub-vector and the weight vector; and construct a prescription generation model based on the prescription vector, the matching function and the weight vector, wherein the weight vector is a preset parameter of the model.

[0067] The processing unit is also used to: obtain predicted prescription information generated by a prescription generation model for multiple patients, and obtain the actual prescription information actually executed corresponding to each predicted prescription information, wherein the actual prescription information includes the patient's actual medication information; obtain the loss function of the prescription generation model; and train the prescription generation model using each predicted prescription information and the corresponding actual prescription information until the threshold of the loss function of the prescription generation model meets the preset conditions, thereby obtaining a target prescription generation model.

[0068] The generating unit is specifically configured to input the at least one target sub-vector into the target prescription generating model to obtain prescription information for the patient's rehabilitation.

[0069] Specifically, the second prediction unit can arbitrarily select and output the impact of lifestyle habits, environmental factors, rehabilitation training, etc. on rehabilitation effects through a multi-output control strategy, wherein the multi-output control strategy is as follows: H(x,W,V,b,c)=σ(xW+b)(xV+c)

[0070] Among them, σ is the activation function, which defaults to the sigmoid function or other functions, which can be selected according to actual needs. x is the second prediction vector, W and V are weight coefficients, and b and c are bias constants.

[0071] In addition, the loss function of the prescription generation model is:

[0072] Among them, x * ,y * Represents predicted prescription information, actual prescription information, is the dense vector (transposed) of the predicted prescription information, x u is the dense vector of predicted prescription information, u is the sequence number of the predicted prescription information vector, y i is the dense vector of actual prescription information, r ui is the regularization term, and γ is the bias coefficient.

[0073] Furthermore, during the model training process, the second prediction unit can also use a reward and punishment system and error feedback method to improve the training efficiency of the model. The reward and punishment system includes:

[0074] Where i and j represent the dimension numbers, N(i) is the predicted prescription information vector, N(j) is the actual prescription information vector, and l is the reward and punishment constant.

[0075] Error feedback methods include:

[0076] Among them, observed t is the actual prescription information, predictedt To predict prescription information, RMSE represents the root mean square error of the system, N is the dimension of the feature vector, and t is the sequence number of the prescription vector.

[0077] In addition, in order to improve the visualization of patient rehabilitation execution management, the processing unit can also observe the patient's execution status through compliance calculation.

[0078] Among them, Task represents the predicted prescription information vector, Execute represents the actual prescription information vector, and Compliance i represents the expected execution capability of the i-th predicted prescription information, and n is the sequence number of the information vector.

[0079] Among them, w i The weight vector representing the i-th predicted prescription information is finally output in the form of a patient performance graph and a comprehensive calculation value of a composite factor.

[0080] In an embodiment of the present application, the first prediction unit, the processing unit, the second prediction unit, and the generation unit can each be one or more processors, controllers, or chips having a communication interface capable of implementing a communication protocol, and may also include a memory and related interfaces, a system transmission bus, etc., if necessary; the processor, controller, or chip executes program-related code to implement corresponding functions. Alternatively, an alternative solution is that the first prediction unit, the processing unit, the second prediction unit, and the generation unit share an integrated chip or a shared processor, controller, memory, or other device. The shared processor, controller, or chip executes program-related code to implement corresponding functions.

[0081] The mobile control module 113 obtains the position of the rehabilitation training patient, measures the distance between the mobile cart 1 and the rehabilitation training patient, controls the mobile cart 1 to move toward the rehabilitation training patient and adjust the height to adapt to rehabilitation training patients of different heights, thereby obtaining a better user experience and providing great convenience for patients with mobility difficulties.

[0082] The rehabilitation training module 114 is used to provide a scenario training game designed based on rehabilitation movements, and the scenario training game is used to implement rehabilitation training for the patient's upper limbs, lower limbs and hands.

[0083] The rehabilitation assessment module 115 is used to assess the motor function of the upper limbs, lower limbs and hands of the rehabilitation training patient. After the rehabilitation training module 114 completes the rehabilitation training, the rehabilitation assessment module 115 assesses the training results in the form of a score.

[0084] Optionally, as shown in Figure 3, the rehabilitation assessment module 115 comprises an assessment assistance unit and an assessment unit. The assessment assistance unit is used to complete the task of identifying rehabilitation assessment movements. The assessment assistance unit is a two-part cascade model, comprising a first classification subunit and a second classification subunit. The assessment unit further comprises an intelligent scale assessment subunit and a compensatory movement quantification assessment subunit.

[0085] The first classification subunit adopts a neural network structure to identify whether the rehabilitation assessment movement the patient is completing is an upper limb arm movement, palm movement, lower limb movement, or upper limb movement. The input of the neural network is a motion feature vector (a vector obtained by threshold judgment of the original signal of the wearable unit), and the output is one of the above four action categories. The neural network contains an input layer, and the number of neurons in this layer is the length of the motion feature vector, which is preset to N. The network contains two hidden layers, the number of neurons in the first hidden layer is 2N, and the number of neurons in the second hidden layer is N. The network contains an output layer with 4 neurons. Finally, the network completes the final output through the softmax layer and one-hot encoding. The network calculation method is as follows: output = one-hot(softmax(z))

[0086] Among them, a j is the hidden layer output, σ is the activation function, N is the number of neurons, i is the loop parameter, w ij is the weight, x i is the input, b j is the bias parameter, z k is the kth output of the output layer, w jk is the weight, b k is the bias parameter, output is the final output, ont-hot is the encoding function, and softmax(z) is the activation function.

[0087] The second classification subunit uses a gated recurrent network to identify specific rehabilitation assessment actions. The input is the original signal of the rehabilitation wearable device, and the output is the specific recognition result. Its calculation method is as follows: g r =σ(W r [h t-1 ,x t ]+b r ) h t ～ =tanh(W h [g r h t-1 ,x t ]+b h ) g z =σ(W z [ht-1 ,x t ]+b z ) h t =(1-g z )h t-1 +g z h t ～

[0088] Where W r ,W z ,W h is the weight parameter, b r ,b h ,b z is the bias parameter, which is automatically optimized by the back propagation algorithm; h t is the state of timestamp t, h t-1 is the state of timestamp t-1, h t ～ is the updated timestamp status; g z Reset gate control vector, g r is the update gate control vector; σ is the activation function, generally used as the Sigmoid function, and tanh is the tanh activation function.

[0089] The intelligent scale evaluation subunit is used to quantitatively score rehabilitation movements. This subunit uses the wearable unit's signal as input and outputs an intelligent quantitative score for the movement. This is accomplished using an LSTM network based on the self-attention mechanism. First, the sub-attention mechanism is used to perform attention encoding on the wearable unit's multivariate signal (x1-xn). The attention mechanism module is calculated as follows:

[0090] Among them, W q is the query matrix, W k is the key matrix, which is automatically learned during the back propagation process of network training; x i and x j is the input feature vector at different time points; AS is the attention score; AW is the attention weight; x j ' is the reconstructed feature vector of a single time point; i, k are loop parameters; exp is the exponential function with e as the base; softmax is the activation function.

[0091] After self-attention encoding, the original signal (x1-x n ) is converted into the reconstructed signal (x1'-x n '), rename it to (z1-z n). At this point, the reconstructed signal is input into the LSTM network to complete the classification task. Assuming that the score of a single action in the original clinical scale is a k-point system (0, 1...k), the LSTM will complete the k-classification task. The calculation of LSTM is as follows: f t =σ(W f ·[h t-1 ,x t ]+b f ) i t =σ(W i ·[h t-1 ,x t ]+b i ) C t ～ =tanh(W C ·[h t-1 ,x t ]+b C ) o t =σ(W o [h t-1 ,x t ]+b o ) h t =o t ×tanh(C t )

[0092] Among them, the input gate acts on the cell state C t (C t-1 is the cell state at the last timestamp) and decides which new information to store in the cell state. The input gate consists of two parts. Input part: constructs input information and decides which information to add as new memory to the cell state. Candidate state part: constructs a candidate cell state C t ～ After constructing these two parts, update the unit state C t The output gate determines the final output h t (h t-1 is the hidden state at the previous timestamp). The output is based on the new cell state C t , which is completed in two steps. The first step is to decide which part of the cell state needs to be output. The second step is to control the final output. f ,W i ,W o is the weight parameter, b f , b i , b c , b o is the bias parameter, which is automatically optimized by the back propagation algorithm; x t is the input feature vector at time t; i t is the input gate, o tis the output gate; σ is the activation function, generally used as Sigmoid function, and tanh is the tanh activation function.

[0093] The output of LSTM will pass through the softmax layer to get the final probability output (p1-p k ), after one-hot encoding, the classification output can be completed, that is, the final intelligent score. The intelligent scale evaluation subunit not only provides the intelligent score of the scale, but also provides the refined scale score s. The calculation method is as follows (i is the loop parameter):

[0094] The compensatory movement quantitative assessment subunit includes compensation detection and compensatory movement quantification. Within the compensatory movement detection model, an internal classifier is used to determine the probability that the movement parameter characteristics are identified as different types of compensatory movements (a total of seven types), and the probability value with the highest probability and the name of the corresponding compensatory movement are used as the detection result. If the detection result shows compensation, the compensatory movement quantitative assessment model is used to determine the quantification threshold interval corresponding to the compensatory movement severity based on the K-nearest neighbor algorithm. The quantification threshold interval corresponding to the quantification threshold interval is then output as the compensatory movement quantitative assessment result.

[0095] The internal classifier for compensatory movement detection is a support vector machine model. First, feature extraction is performed on the limb movement data based on the completeness, smoothness, and degree of fluctuation of the target patient's limb movement test. Motion parameter features related to the target patient's trunk, shoulder, and elbow joints are screened to generate motion feature vectors. To detect the seven types of compensatory movements present in the target patient, seven binary classification models are constructed as representatives of the compensatory movements in that category. Finally, the outputs of the seven SVM binary classifiers are integrated to detect and classify multiple compensatory movements. Specifically, each binary classification model takes a feature vector as input and outputs probabilities p0 and p1 of classification as 0 or 1. If p1 is greater than p0, the final compensatory movement type is determined by the model with the largest p1 output from each of the seven models. If p1 is not greater than p0, no compensation is present.

[0096] The compensatory movement quantitative assessment model uses the quantitative threshold interval of the compensatory movement severity as the assessment label of the quantitative category, determines the quantitative threshold interval corresponding to the movement parameter characteristics based on the K-nearest neighbor algorithm, and outputs the quantitative category corresponding to the quantitative threshold interval as the compensatory movement quantitative assessment result.

[0097] Specifically, the quantitative threshold intervals for compensatory movement severity refer to the classification of the same compensatory movement into K levels based on its severity. The range of each severity level is the quantitative threshold interval for compensatory movement. The quantitative threshold intervals for mild, moderate, and severe compensation are determined by clustering the movement parameter characteristics of multiple stroke patients with these types of compensatory movements using a K-clustering algorithm. Cluster analysis based on a large sample of data using the K-means clustering algorithm yields labels for the compensatory movement severity classification intervals. K-level compensatory movement severity assessment thresholds are then determined by parameter setting, where K is greater than or equal to 3. When K is 3, the quantitative threshold intervals corresponding to mild, moderate, and severe compensation are the first, second, and third threshold intervals, respectively. Subsequently, a supervised machine learning algorithm is used to divide the training set into a test set. The movement parameter characteristics and multi-level assessment labels are used as input to the compensatory movement quantitative assessment model. A K-nearest neighbor machine learning algorithm is then used to quantitatively assess the severity of compensation based on the multi-level threshold labels.

[0098] In an embodiment of the present application, the first prediction unit, the processing unit, the second prediction unit, the generation unit, the evaluation auxiliary unit, the evaluation unit, the intelligent scale evaluation subunit, the compensatory movement quantitative assessment subunit, the first classification subunit, and the second classification subunit can respectively be one or more processors, controllers, or chips with a communication interface capable of implementing a communication protocol, and may also include a memory and related interfaces, a system transmission bus, etc. if necessary; the processor, controller, or chip executes program-related codes to implement corresponding functions. Alternatively, an alternative solution is that the first prediction unit, the processing unit, the second prediction unit, the generation unit, the evaluation auxiliary unit, the evaluation unit, the intelligent scale evaluation subunit, the compensatory movement quantitative assessment subunit, the first classification subunit, and the second classification subunit share an integrated chip or a shared processor, controller, memory, or other device. The shared processor, controller, or chip executes program-related codes to implement corresponding functions.

[0099] As shown in FIG4 , the mobile unit 12 includes rollers 123, electric push rods 122, and a driver 121 disposed on the mobile cart 1. The driver 121 is connected to the mobile control module 113. The mobile control module 113 includes a controller for controlling the driver 121 to move the rollers 123 and control the electric push rods 122 to extend and retract, thereby adjusting the height of the embedded host computer 11 to adapt its height to the current rehabilitation training patient.

[0100] In this embodiment, the embedded host computer 11 adopts an SK-27A model all-in-one computer, which adopts a 27-inch capacitive touch screen, a Core i5-3210 CPU, a Longsys 4G memory, a Longsys 128G solid-state hard drive, and the rest includes a power adapter, a connecting cable, a stylus, etc. The all-in-one computer is installed on the upper part of the mobile cart 1 body in an embedded manner. The mobile cart 1 also includes a base 101 and a connecting block 102, as shown in Figure 2. The roller 123 is set at the bottom of the base 101, and the electric push rod 122 is set in the connecting block 102. The embedded host computer 11 is connected to the output end of the electric push rod 122, so that the height of the embedded host computer 11 can be controlled by the electric push rod 122. The electric push rod uses a TJC-C4 model electric push rod.

[0101] It should be noted that in this embodiment, the rehabilitation training module 114 and the rehabilitation assessment module 115 constitute a rehabilitation training and assessment system. After logging into the rehabilitation training and assessment system, you can select rehabilitation training and rehabilitation assessment content as needed. If you choose rehabilitation training content, you need to put on the rehabilitation training gloves 201 and the somatosensory instrument 202 first, and then select the rehabilitation training function according to the display window. The rehabilitation training function is loaded with many rehabilitation training games. Rehabilitation training patients can choose the difficulty of the corresponding rehabilitation training game according to their own situation. For patients who use this device for the first time, the system is also loaded with teaching videos. Rehabilitation training patients can learn how to use it according to the video, which can help users better use the device and further improve the user experience. Rehabilitation training games use fun games that are close to life. Through games, the patient's upper limbs, lower limbs and fingers can be fully trained to speed up the patient's rehabilitation progress.

[0102] As shown in Figure 5, the rehabilitation training glove 201 comprises a glove body and a first housing mounted on the glove body. The first housing houses a first battery, a first circuit board, and a bend sensor. A first indicator light is also provided on the first housing. In this embodiment, the glove body is made of nylon, and the cuffs are elasticated for enhanced comfort. The gloves are available in various sizes to suit a variety of users. The first battery is a Zhongshun 602035 lithium battery, which provides power to the remaining components of the glove. The bend sensor is a FS-L-0055-253-ST bend sensor, which measures data from the rehabilitation patient during training. The first circuit board transmits the bend sensor data to the data acquisition module 111 via an integrated wireless transmitter module. The first circuit board also controls the flashing of the first indicator light. When the corresponding channel device of the rehabilitation training glove 201 is operating normally, the first circuit board's control indicator light is green (or flashing); when the corresponding channel device is receiving no data, the control indicator light is gray; and when the corresponding channel device is receiving abnormal data, the control indicator light is red.

[0103] As shown in Figure 6, the somatosensory instrument 202 includes a strap, a second shell installed on the strap, a second circuit board and a second battery are provided in the second shell, and a second indicator light is provided on the second shell. In this embodiment, the number of somatosensory instruments 202 can be multiple, which are mainly used to collect upper limb and lower limb movement data. The strap can be selected in different sizes as needed, and is used to fix the second shell and the second circuit board, second battery, etc. therein in corresponding positions. The second battery can be of the same model as the first battery and is used to provide power support. Corresponding sensors connected to the second circuit board are provided in the second shell as needed, such as a posture sensor. The second circuit board is used to control the flashing of the second indicator light, and is also used to send the sensor data to the data acquisition module 111 through the integrated wireless transmission module.

[0104] It can be understood that in this embodiment, in addition to the rehabilitation training gloves 201 and the somatosensory device 202, other wearable devices or rehabilitation training devices, such as a treadmill, a rehabilitation training helmet, etc., can also be used.

[0105] Preferably, the data acquisition module 111 is connected to a receiver 1111 and a camera 1112. The receiver 1111 is wirelessly connected to the rehabilitation training glove 201 and the somatosensory device 202, and is used to receive motion data from the rehabilitation training glove 201, the somatosensory device 202, and other devices. The camera 1112 captures the movements of the rehabilitation patient to achieve motion data collection. In this embodiment, the camera 1112 can be the camera 1112 built into the all-in-one computer, or a separate camera 1112 can be connected. There are no special requirements for the model of the camera 1112, as long as it is compatible with the all-in-one computer and the rehabilitation training and assessment system.

[0106] Preferably, in this embodiment, the all-in-one computer can be connected to a cloud server via the Internet. The cloud server stores exercise prescriptions, and exercise prescriptions can be downloaded or training data can be uploaded through the cloud server. In actual applications, rehabilitation patients can also connect to terminal devices such as hospitals through the cloud server, enabling remote online guidance and diagnosis by doctors.

[0107] As shown in Figure 4, the device also includes a human-computer interaction column 13, which is connected to the computer-integrated device via a wired or wireless connection. A card reader area and a keypad area are provided on the human-computer interaction column 13. The card reader area is used to identify the rehabilitation patient, and the keypad area is provided with multiple keys that assist in playing scenario-based training games. Before using the device for rehabilitation training, the rehabilitation patient needs to swipe their ID card or identification card in the card reader area. The computer-integrated device then authenticates the patient using the ID card or identification card. The patient can then access the rehabilitation training and assessment system and play the rehabilitation training game by pressing the keys in the keypad area. It is understood that a management system for rehabilitation patients can be established based on the human-computer interaction column 13, with files created for each rehabilitation training and assessment of the patient, allowing for tracking of rehabilitation progress at any time. This not only improves the user experience, but also reduces the burden on staff through detailed management files, making the entire rehabilitation training process more standardized.

[0108] Preferably, the device also includes a power module 14, which includes a magnetic ring, a transformer, a filter, a data line, etc., for providing power and processing power signals. The magnetic ring is selected from V18004, V18005 or V18007 models to achieve electromagnetic shielding. The filter is selected from RSEN-2006L model EMC filter to process power signals. The transformer is selected from H0128-823-0250 model isolation transformer. The magnetic ring, transformer, and filter are all arranged in a cavity within the base 101, which makes the base 101 heavier and facilitates maintaining the stability of the embedded host computer 11 connected thereto during movement. The power supply and the embedded host computer 11 are connected between the components as needed, which will not be elaborated in detail here.

[0109] As shown in Figure 7, the motion control module 113 is connected to a radar detector for detecting the position of the rehabilitation patient. The motion control module 113 is also used to control the operation of the driver 121 and the electric push rod 122. In this embodiment, the radar detector can be a SENKYLASER SK-C102D laser scanning radar. The motion control module 113 uses the Gmapping algorithm for autonomous positioning and mapping. It is understood that the device also includes a GPS positioning module for locating the current position of the mobile cart 1. The radar detector then scans the surrounding environment and determines the location of the rehabilitation patient. Ultimately, the motion control module 113 controls the rollers 123 via the driver 121 to move in the direction of the rehabilitation patient. In this embodiment, there are four rollers 123, which are circumferentially arranged on the bottom of the mobile cart 1. The driver 121 uses two motors, one of which is connected to two rollers 123. The controller controls the forward and reverse rotation of the motor to drive the rollers 123, thereby achieving forward and backward movement of the mobile cart 1. Another motor is connected to the other two rollers 123. The controller controls the forward and reverse rotation of the other motor to drive the other two rollers 123, thereby turning the cart 1 left and right. Controlling the forward, backward, left, and right movement of the device using motors is conventional technology and is similar to controlling the forward, backward, left, and right movement of a toy car using a handle. The controller can be a remote control. The remote control and motor are connected via Wi-Fi. The remote control is equipped with forward, backward, left, and right buttons, which control the forward, backward, left, and right movement of the device.

[0110] It should be noted that the driver 121 uses a servo motor as a driver, and its model can be HC-KFS, HC-MFS, HC-SFS, HC-RFS and HC-UFS series. The output end of the driver 121 is connected to the roller 123, and the driver 121 is controlled by the motion control module 113 to realize the movement of the cart 1. In order to better realize the forward, backward, left and right movement of the cart 1, there are four rollers 123 and two drivers 121. The four rollers 123 are set at the bottom of the base 101. One driver 121 is connected to two rollers 123. By controlling the forward and reverse rotation of the driver 121, the cart 1 is moved forward and backward. Another driver 121 is connected to the other two rollers 123. By controlling the forward and reverse rotation of the driver 121, the cart 1 is turned left and right.

[0111] Since different users have different heights, in order to achieve automatic height adjustment of the device, an infrared sensor connected to the mobile control module 113 is set on the top of the mobile cart 1. The patient only needs to wear a corresponding infrared signal transmitter when using it. At the beginning of use, the mobile control module 113 controls the electric push rod 122 to operate within a certain range, so that the embedded host computer 11 moves within a certain height range. When the infrared sensor and the infrared signal transmitter are at the same height, the infrared sensor will receive the infrared signal emitted by the infrared signal transmitter. At this time, the mobile control module 113 controls the electric push rod 122 to stop working, and the height that is most suitable for the current patient is obtained.

[0112] In other embodiments, the device can realize manual adjustment of the height. The lifting program of the electric push rod 122 is integrated into the movement control module 113. By opening the lifting control interface of the electric push rod 122 of the movement control module 113 and clicking the raising or lowering option, the embedded host computer 11 can be raised or lowered.

[0113] In other embodiments, the mobile control module 113 includes a remote control, which is connected to the embedded host computer 11 via WIFI. The mobile control module 113 controls the electric push rod 122 through the corresponding up or down buttons in the remote control, thereby achieving the raising or lowering of the embedded host computer 11.

[0114] In this embodiment, data acquisition module 111 collects raw limb motion signals from rehabilitation patients via receiver 1111 and camera 1112 and sends them to data processing module 112 for processing. Data processing module 112 first preprocesses the limb motion signals to eliminate the interference and influence of system noise. Next, feature extraction is performed on the raw limb motion signals. In this embodiment, feature extraction is performed based on the degree of completion and smoothness of the patient's movements, resulting in a one-dimensional feature vector for the rehabilitation training movements. Finally, a multi-source fusion algorithm is used to quantitatively evaluate features across multiple dimensions. Based on the results, tasks and characters in the virtual reality scene are driven (controlled) and human-computer interaction feedback is provided.

[0115] The rehabilitation assessment module 115 establishes a motion recognition model for wearable devices based on a time series network. After completing the rehabilitation training movements, the patient needs to evaluate the quality of the training movements. After obtaining the one-dimensional feature vector of the rehabilitation training movements processed by the data acquisition module 111, an LSTM scale assessment network based on an attention mechanism and a compensation detection network based on machine learning are used to establish a training movement assessment model. In the movement assessment, the prediction process is the evaluation process (giving movement scores and compensation scores), and the data is included in the training set, and the training model is updated so that the model has a self-learning function. On the basis of this model, nonlinear dynamic tools such as Poincare difference scatter plots, maximizing Lyapunov exponents, and distribution entropy are used to design visual quantitative assessment parameters, so that the quantitative assessment model can give visual quantitative scores for the training movements.

[0116] The above is merely the preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application shall be included in the scope of protection of this application. Therefore, the scope of protection of this application shall be based on the scope of protection of the claims.

Claims

1. An intelligent rehabilitation device, characterized in that: The device comprises: a mobile cart; The mobile cart is provided with an embedded host computer, a mobile unit and a wearable unit, and both the mobile unit and the wearable unit are communicatively connected with the embedded host computer; The embedded host computer includes a rehabilitation training module, a rehabilitation assessment module, a data acquisition module, a data processing module, a movement control module and a storage module; The rehabilitation training module is used to provide a scenario training game designed based on rehabilitation movements, and the scenario training game is used to implement the training of the upper limbs, lower limbs and hands of the rehabilitation training patient; The rehabilitation assessment module is used to assess the motor functions of the upper limbs, lower limbs and hands of the rehabilitation training patient; The data collection module is used to collect information about the rehabilitation training patient, the information includes feature vectors corresponding to multiple dimensions of the patient, each feature vector is used to indicate the patient information under each dimension, and the dimensions include: patient medical history files, living habits, environmental factors and rehabilitation training status; The data processing module is used to predict the degree of influence of the patient information of each dimension on the patient's rehabilitation based on the feature vector of each dimension, obtain a first prediction vector corresponding to the feature vector of each dimension, the first prediction vector is used to indicate the degree of influence of the patient information of each dimension on the patient's rehabilitation, and perform feature splicing on each of the first prediction vectors to obtain a target vector; predict the rehabilitation effect corresponding to the patient information of each dimension based on the target vector to obtain a second prediction vector, the second prediction vector is used to indicate the rehabilitation effect corresponding to the patient information of each dimension; generate prescription information for the patient's rehabilitation based on the second prediction vector, the prescription information including medication information and drug dosage; The mobile control module controls the mobile cart to move and adjust its height by acquiring the position of the rehabilitation training patient and measuring the distance between the mobile cart and the rehabilitation training patient; The mobile unit includes a roller, an electric push rod and a driver arranged on the mobile cart, the driver is connected to the mobile control module, and the mobile control module drives the roller to move and adjust the position and controls the electric push rod to adjust the height through the driver; The wearable unit includes a rehabilitation training glove and a somatosensory instrument. Both the rehabilitation training glove and the somatosensory instrument perform rehabilitation training through the rehabilitation training module, and send rehabilitation training data to the data acquisition module.

2. The device according to claim 1, characterized in that The data acquisition module includes a receiver and a camera, and the receiver is wirelessly connected to the rehabilitation training glove and the body sensor respectively; The receiver is used to receive the motion data of the rehabilitation training gloves and the somatosensory instrument, and the camera collects the motion data by capturing the movements of the rehabilitation training patient; The data processing module is also used to process the motion data based on multi-source information fusion.

3. The device according to claim 1, characterized in that The device also includes: A human-computer interaction column, wherein the human-computer interaction column is provided with a card reading area and a button area, wherein the card reading area is used to identify the rehabilitation training patient, and the button area is provided with a plurality of buttons, wherein the buttons are used to assist in performing the scenario training game; A power module, comprising a magnetic ring, a transformer, a filter, and a data line, and is used to provide power and process power signals; The embedded host computer is also connected to the cloud server in communication; The cloud server stores exercise prescriptions; The embedded host computer is also used to download the exercise prescription or upload the training data through the cloud server.

4. The device according to claim 1, characterized in that The rehabilitation training glove comprises a glove body and a first shell mounted on the glove body; A first battery, a first circuit board and a bending sensor are arranged in the first shell, and a first indicator light is arranged on the first shell.

5. The device according to claim 1, characterized in that The somatosensory instrument comprises a strap and a second shell mounted on the strap; A second circuit board and a second battery are arranged in the second shell, and a second indicator light is arranged on the second shell.

6. The device according to claim 1, characterized in that The rehabilitation assessment module includes an assessment auxiliary unit and an assessment unit, wherein the assessment unit includes an intelligent scale assessment subunit and a compensatory movement quantitative assessment subunit; The assessment auxiliary unit is used to complete the task of recognizing the rehabilitation assessment action and obtain the recognition result; The evaluation unit is used to obtain a corresponding evaluation result based on the recognition result.

7. The device according to claim 1, characterized in that The embedded host computer realizes terminal interconnection through a network for remote medical diagnosis or online rehabilitation training guidance.

8. The device according to claim 1, characterized in that The mobile control module is connected with the radar detector and the positioning module; The radar detector is used to detect the position of the rehabilitation training patient; The positioning module is used to determine the current position of the mobile cart.

9. The device according to any one of claims 1 to 8, characterized in that: The mobile cart includes a base and a connecting block; The roller is arranged at the bottom of the base, the electric push rod is arranged in the connecting block, and the embedded upper computer is connected to the top of the electric push rod.

10. The device according to claim 1, characterized in that The mobile control module is connected to an infrared sensor and an infrared signal transmitter. When the device is in use, position and height adjustment are achieved through the infrared sensor, the infrared signal transmitter and the mobile control module.

Images