A multi-modal exoskeleton rehabilitation method and system

Abstract

The application discloses a kind of multi-mode exoskeleton rehabilitation method and system, belong to rehabilitation training technical field.Method includes: obtaining the data of patient's Fugl-Meyer Assessment Scale, Berg Balance Scale and Barthel Index;Stroke rehabilitation prescription model is input, and initial exoskeleton rehabilitation intervention scheme is generated;According to scheme selection module type and passive gait, power, active or resistance mode are implemented training;Gait data is collected in real time during training, and update scheme is generated by rehabilitation capacity evaluation model;After meeting preset condition, scheme is automatically switched.The application realizes rehabilitation prescription autonomous decision based on clinical general scale, covers whole rehabilitation cycle through cross-device multi-mode intervention, and forms dynamic closed-loop evaluation and real-time updating mechanism, effectively improves the accuracy, efficiency and safety of rehabilitation training.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation training technology, specifically to a data-driven multimodal exoskeleton rehabilitation method and system, which is particularly suitable for rehabilitation assistance for patients with lower limb motor dysfunction after stroke. Background Technology

[0002] Stroke often leads to severe impairment of lower limb motor function. Scientific and timely rehabilitation training can effectively promote neurological compensation and reorganization, improve motor control ability, and help rebuild walking function.

[0003] Currently, rehabilitation therapy in my country still primarily relies on traditional, manual training models. For patients with functional motor disorders such as hemiplegia, "one-on-one" or "many-to-one" assisted learning methods are commonly used: rehabilitation therapists develop training plans based on assessment scales and utilize equipment such as standing frames and parallel bars for training. However, my country faces a severe shortage of rehabilitation medical resources and professional rehabilitation personnel, resulting in many patients missing the optimal intervention period due to the inability to access timely professional rehabilitation services. Furthermore, rehabilitation outcomes are highly dependent on the therapist's individual experience, making it difficult to guarantee training quality in resource-scarce areas.

[0004] Currently, some technologies have attempted to introduce exoskeleton robots into rehabilitation training, but the following shortcomings still exist: 1. Regarding autonomous decision-making in rehabilitation programs: While existing technologies involve assessment and rehabilitation program development based on patient gait data, they generally suffer from unclear correspondence between assessment benchmarks and rehabilitation programs, and poor model applicability. For example, although patent CN202511352066.1 utilizes gait quantitative analysis for dynamic assessment of rehabilitation parameters, it does not clearly define the design of a specific rehabilitation program decision-making model, and its multidimensional gait scoring model is not yet widely used in clinical practice. There is a lack of a universally applicable and easy-to-operate autonomous decision-making mechanism for rehabilitation programs based on a clinically common scale.

[0005] 2. Regarding multimodal rehabilitation interventions using exoskeletons: Most existing technologies only provide a single type of rehabilitation mode or are limited to a single device. For example, patent CN202111676980.3 involves multimodal training based on sensor assessment, but requires the patient to undergo initial training before a plan can be determined; patent CN201911183996.3 utilizes electromyography (EMG) signals for multi-stage training, but requires prior multi-channel EMG signal acquisition. These solutions generally fail to achieve full-cycle, multimodal coverage across rehabilitation exoskeleton devices, making it difficult to adapt to the continuously changing needs of patients from early passive intervention to late-stage active resistance training.

[0006] 3. Regarding online assessment and rehabilitation program updates for exoskeletons: Existing technologies lack real-time online assessment capabilities during exoskeleton training. For example, patent CN202511406792.7 can dynamically assess joint range of motion and recommend training modes, but it is limited to a few modes of heavy lower limb exoskeletons and cannot provide further assistance to patients with independent walking ability in the late stages of rehabilitation. Overall, existing solutions cannot utilize the exoskeleton's own sensors to determine the patient's rehabilitation fit in real time and update the optimal program in a timely manner, resulting in a lag in rehabilitation programs and a heavy reliance on the therapist's continuous attention and subjective judgment.

[0007] Therefore, how to provide an exoskeleton rehabilitation method that is based on clinically universal scales for autonomous decision-making, covers multimodal interventions throughout the entire rehabilitation cycle, and has the ability to conduct real-time online assessments and update plans has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] The present invention aims to solve the above-mentioned problems in the prior art and provide a data-driven multimodal exoskeleton rehabilitation method, which realizes intelligent autonomous decision-making for rehabilitation prescriptions, flexible switching of multimodal interventions across devices, and dynamic evaluation and plan updates during the rehabilitation process.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A multimodal exoskeleton rehabilitation method includes the following steps: Step S1: Obtain patient clinical assessment scale data, including Fugue-Meier rating scale data, Berg balance scale data, and Barthel Index data; Step S2: Input the scale data into the pre-trained stroke rehabilitation prescription model, and the rehabilitation prescription model outputs an initial exoskeleton rehabilitation intervention plan that matches the patient's current rehabilitation ability; wherein, the rehabilitation prescription model is trained based on a high-quality rehabilitation dataset, and the high-quality rehabilitation dataset consists of the scale data and rehabilitation intervention mode labels manually annotated by physicians; Step S3: Based on the initial exoskeleton rehabilitation intervention plan, select the corresponding exoskeleton type and rehabilitation intervention mode from the modular exoskeleton system, and implement rehabilitation training for the patient; wherein, the exoskeleton type includes lower limb rehabilitation exoskeleton suitable for the early and middle stages of rehabilitation, and portable walking assistive exoskeleton suitable for the middle and late stages of rehabilitation; the rehabilitation intervention mode includes passive gait mode, assisted mode, active mode and resistance mode. Step S4: During rehabilitation training, gait-related data of the patient is collected in real time through the built-in sensors of the exoskeleton, and the gait-related data is input into a pre-trained rehabilitation ability assessment model to generate an updated exoskeleton rehabilitation intervention plan; wherein, the rehabilitation ability assessment model is trained based on the preset exoskeleton data collection dataset and the rehabilitation intervention mode updated labels manually labeled by the physician. Step S5: Determine whether the conditions for updating the treatment plan are met. If they are met, replace the current treatment plan with the updated exoskeleton rehabilitation intervention plan and continue to provide rehabilitation training to the patient.

[0010] Furthermore, the stroke rehabilitation prescription model in step S2 adopts an attention-enhanced random forest model. The attention-enhanced random forest model assigns differentiated weights to each assessment dimension in the Fugue-Meier rating scale data, the Berg balance scale data, and the Barthel Index data through an attention mechanism, so as to capture the correlation weights between different scale characteristics and rehabilitation intervention programs.

[0011] Furthermore, the passive gait mode in step S3 specifically involves using a position control mode on the lower limb rehabilitation exoskeleton to drive the exoskeleton joints to move the patient's limbs strictly according to a preset standard gait trajectory in order to correct abnormal gait.

[0012] Furthermore, the assist mode in step S3 is specifically as follows: the torque sensor built into the exoskeleton joint collects the human-machine interaction torque in real time, and after gravity compensation, it is input into the admittance controller. The admittance controller dynamically and flexibly corrects the standard gait trajectory according to the preset virtual stiffness coefficient, damping coefficient and inertia coefficient, and generates a personalized assist gait trajectory that fits the patient's real-time force exertion state.

[0013] Furthermore, the active mode in step S3 specifically refers to the exoskeleton adopting a zero-torque following control mode. Through gravity compensation, joint friction loss compensation, and inertia compensation, the output torque of the drive motor is adjusted to approach zero, and the exoskeleton follows the patient's autonomous movement trajectory to achieve imperceptible zero-force following. At the same time, gait data is monitored in real time and safety protection is triggered.

[0014] Furthermore, the resistance mode in step S3 is specifically as follows: the exoskeleton adopts an impedance control mode to establish a dynamic relationship between joint torque and position and velocity, and provides adjustable damping resistance opposite to the direction of movement during the patient's active movement; the impedance control mode is based on the error between the feedback angle and the desired angle of the hip joint exoskeleton, as well as the error between the feedback angular velocity and the desired angular velocity, and calculates the output control torque as the resistance torque through virtual stiffness term and virtual damping term.

[0015] Furthermore, in step S4, the exoskeleton's built-in sensors include joint end encoders, torque sensors, and inertial measurement units, and the collected gait-related data include joint angles, angular velocities, gait cycle phases, and human-machine interaction torques.

[0016] Furthermore, in step S5, determining whether the scheme update conditions are met requires that the following sub-conditions be met simultaneously: Sub-condition 1: The updated exoskeleton rehabilitation intervention program differs from the current program; Sub-condition 2: The update scheme generated in n consecutive times remains unchanged, where n is a preset positive integer; Sub-condition 3: An updated plan is generated for each complete rehabilitation training session.

[0017] The present invention also provides a multimodal exoskeleton rehabilitation system, comprising: The data acquisition module is used to acquire patient clinical assessment scale data, including Fugue-Meier rating scale data, Berg balance scale data, and Barthel Index data. The prescription generation module is used to input the scale data into a pre-trained stroke rehabilitation prescription model and output an initial exoskeleton rehabilitation intervention plan that matches the patient's current rehabilitation ability. The modular exoskeleton execution module is used to select the exoskeleton type and rehabilitation intervention mode according to the initial exoskeleton rehabilitation intervention plan to carry out rehabilitation training for the patient. The exoskeleton type includes a lower limb rehabilitation exoskeleton and a portable walking assistive exoskeleton. The rehabilitation intervention mode includes a passive gait mode, an assistive mode, an active mode, and an resistance mode. The online assessment and update module is used to collect gait data through the exoskeleton's built-in sensors during training and input it into the rehabilitation ability assessment model to generate an updated exoskeleton rehabilitation intervention plan. The scheme switching module is used to replace the current scheme with the updated scheme when preset update conditions are met.

[0018] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the multimodal exoskeleton rehabilitation method as described above.

[0019] Beneficial effects

[0020] Compared with the prior art, the present invention has the following beneficial effects: Automated and precise matching of treatment plans: This invention only requires input of clinically common data from the Fuqua-Meyer, Berger's Balance, and Barthel Index scales to automatically generate personalized exoskeleton rehabilitation intervention plans through a pre-trained stroke rehabilitation prescription model. The model employs an attention-enhanced random forest structure, which can accurately capture the correlation weights between different scale features and rehabilitation plans. This solves the problems of traditional rehabilitation plans relying on physician experience, cumbersome procedures, and low matching accuracy, achieving intelligent, standardized, and precise rehabilitation prescriptions.

[0021] Full-cycle, multi-modal rehabilitation coverage: This invention, through modular exoskeleton design, provides two types: lower limb rehabilitation exoskeleton and portable assistive walking exoskeleton, integrating four intervention modes: passive gait, assisted gait, active gait, and resistance gait. From passive correction in the early stage of rehabilitation, to assisted guidance in the middle stage, and then to active following and resistance reinforcement in the final stage, it achieves seamless coverage of the entire rehabilitation cycle for stroke patients, breaking through the technical limitations of single devices and single modes.

[0022] Dynamic Closed-Loop Assessment and Real-Time Updates: This invention constructs a dynamic closed-loop mechanism of "intervention-assessment-update." While the exoskeleton performs rehabilitation training, built-in sensors collect gait data in real time. This data is analyzed online by a rehabilitation capability assessment model, which generates an updated plan and automatically switches when preset conditions are met. This method effectively avoids the lag in updates found in traditional rehabilitation plans, ensuring that the intervention intensity remains dynamically adapted to the patient's real-time functional state, significantly improving rehabilitation efficiency and safety.

[0023] High applicability and scalability: The input data relied upon by this invention are routine hospital assessment scales, without relying on special sensors or complex preprocessing; the decision model can be encapsulated and deployed in the exoskeleton control system, without relying on the physician's personal experience. This makes this method easy to promote and apply in rehabilitation institutions at different levels, helping to alleviate the contradiction of uneven distribution of rehabilitation resources and improve the homogeneity of overall rehabilitation services. Attached Figure Description

[0024] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 Examples of high-quality patient datasets and rehabilitation program labels provided in the embodiments of this application; Figure 2 A schematic diagram of the autonomous decision-making model for the rehabilitation program provided in this application; Figure 3 A schematic diagram of the modular exoskeleton structure provided in this application; Figure 4 The logic diagram of the passive gait pattern of the lower limb rehabilitation exoskeleton provided in this application; Figure 5 The logic diagram of the exoskeleton assistance mode based on admittance control provided in this application; Figure 6 The active mode logic diagram of the exoskeleton based on zero-force follow control provided in this application; Figure 7 The logic diagram of the resistance training mode of exoskeleton based on impedance control provided in this application; Figure 8 A schematic diagram of a patient-scale-driven multimodal rehabilitation program across exoskeleton devices provided for this application; Figure 9 A schematic diagram illustrating the logic of self-development and updating of the exoskeleton rehabilitation program provided in this application; Figure 10 This application provides a multimodal exoskeleton rehabilitation method with corresponding computer-readable storage media system framework diagram. Detailed Implementation

[0026] Detailed embodiments of the invention are disclosed herein as needed; however, it should be understood that the disclosed embodiments are merely illustrative of the invention, and the invention may be implemented in different and alternative forms. The drawings are not necessarily drawn to scale, and certain features may be exaggerated or reduced to show details of specific components. Therefore, the specific structural and functional details disclosed herein should not be construed as limiting, but rather as a representative basis to teach those skilled in the art to employ the invention differently. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of this invention.

[0027] Example 1: A Multimodal Exoskeleton Rehabilitation Method This embodiment provides a data-driven, multimodal exoskeleton rehabilitation method, the specific process of which is as follows: Step S1: Obtain patient clinical assessment scale data First, routine clinical functional assessments were performed on stroke patients, and data from the Fugl-Meyer Assessment (FMA), Berg Balance Scale (BBS), and Barthel Index (BI) were obtained. The FMA was used to assess the degree of lower limb motor function recovery, the BBS to assess sitting and standing balance, and the BI to assess daily living abilities. These scale data were all structured data routinely collected by the hospital's rehabilitation department, requiring no additional testing.

[0028] Step S2: Generate an initial rehabilitation intervention plan using a stroke rehabilitation prescription model. The scale data obtained in step S1 is input into the pre-trained stroke rehabilitation prescription model. The training process of this model is as follows: First, construct a high-quality rehabilitation dataset. For example... Figure 1 As shown, the dataset consists of two parts: one part derived from clinically collected stroke patient scale data and corresponding rehabilitation intervention programs, and the other part from artificially simulated data that conforms to clinical distribution characteristics. Both parts of the data were manually labeled by rehabilitation physicians with extensive clinical experience, forming a mapping label of "scale data - rehabilitation intervention mode". The rehabilitation intervention modes include six specific programs: lower limb rehabilitation exoskeleton passive gait mode, lower limb rehabilitation exoskeleton assisted mode, assistive walking exoskeleton assisted mode, assistive walking exoskeleton active mode, and assistive walking exoskeleton rehabilitation training mode (i.e., resistance mode).

[0029] Secondly, an attention-enhanced random forest model was selected as the classification model for training. For example... Figure 2 As shown, during the model training phase, the input consists of scores for various assessment dimensions of the Fuqua-Meyer Assessment Scale, the Berger Balance Scale, and the Barthel Index. The output is the classification results for the corresponding six exoskeleton rehabilitation intervention modes. This attention-enhanced random forest model introduces an attention mechanism to automatically learn and assign differentiated weights to features of different scales and assessment dimensions during the construction of the decision tree. For example, the model assigns higher weights to core features that determine whether to use the passive gait mode—such as the lower limb dissociation score in the Fuqua-Meyer Scale and the standing balance score in the Berger Balance Scale; and lower weights to secondary features that are less associated with the choice of rehabilitation mode. This mechanism effectively solves the problem of inaccurate feature extraction from clinical scale data in traditional random forests, improving the accuracy of multi-classification decisions for stroke rehabilitation programs.

[0030] In practical application, the scale data of the patient to be assessed is input into the trained model, which then automatically outputs an initial exoskeleton rehabilitation intervention plan that matches the patient's current rehabilitation ability. This plan specifically specifies the type of exoskeleton to be used (lower limb rehabilitation exoskeleton or portable walking exoskeleton) and the corresponding intervention mode.

[0031] Traditional random forests uniformly assign feature weights to clinical scale data, failing to accurately identify the correlation weights between core features such as "joint movement function" in the Fugl-Meyer scale, "balance stability" in the Berg balance scale, and "ability to perform daily activities" in the Barthel Index, and rehabilitation plans. This leads to feature redundancy and the masking of key information, resulting in insufficient model decision accuracy. SVM models are prone to overfitting when processing high-dimensional, small-sample scale data and are highly sensitive to noisy data (such as random errors in patient assessments), exhibiting poor generalization ability and failing to adapt to scale data scenarios involving multi-center, multi-type stroke patients. Conventional neural networks (such as CNNs and MLPs) require large amounts of labeled data for training, while clinical scale data suffers from high labeling costs and limited sample sizes, easily leading to insufficient model training and large decision biases. Furthermore, the decision-making process of neural network models is "black box," unable to explain the logical correlation between scale features and rehabilitation plans, failing to meet the medical field's requirement for interpretable decision-making. The AERF model of this invention introduces an attention mechanism, which can automatically learn and assign differentiated weights to features of different scales and assessment dimensions. For example, it assigns higher weights to core features influencing exoskeleton mode selection (such as the lower limb dissociation score of the Fugl-Meyer scale and the balance score of the Berg scale) and lower weights to secondary features, effectively addressing the shortcomings of inaccurate feature extraction in traditional random forests. Simultaneously, the attention mechanism optimizes the splitting strategy of the decision tree, reducing redundant decision nodes and improving model training efficiency and decision-making speed. Furthermore, the AERF model filters noisy features in the scale data through the attention mechanism, reducing the impact of random errors on model training. Combined with the ensemble learning characteristics of random forests, it enhances the model's generalization and anti-interference capabilities, effectively adapting to scale data in different hospitals and assessment scenarios, solving the problems of insufficient generalization and overfitting in conventional classification models. Moreover, the AERF model balances the stability of ensemble learning with the interpretability of the attention mechanism, requiring no massive amounts of labeled data; effective training can be completed using only high-quality clinical scale datasets, and it can clearly output the influence weights of each scale feature on rehabilitation plan decisions.

[0032] Step S3: Implement multimodal exoskeleton rehabilitation training according to the initial plan. Based on the initial plan generated in step S2, select the appropriate exoskeleton type and rehabilitation intervention mode from the modular exoskeleton system, and implement rehabilitation training for the patient.

[0033] The modular exoskeleton structure used in this embodiment is as follows: Figure 3 As shown. Figure 3The exoskeleton shown on the left is suitable for lower limb rehabilitation in the early and middle stages of rehabilitation. It mainly consists of a support frame 1, a hip joint module 2, a knee joint module 3, and an ankle joint module 4. It can provide comprehensive torque assistance to the patient's hip, knee, and ankle joints to achieve gait correction. Figure 3 The right side shows a portable walking exoskeleton suitable for the middle and late stages of rehabilitation. Taking the hip joint walking exoskeleton as an example, its structure mainly consists of a control backpack 11, a hip joint module 12, and a thigh bar module 13. The hip joint module, knee joint module, and ankle joint module can all be detached from the lower limb rehabilitation exoskeleton and controlled independently for use as a portable walking exoskeleton. This modular design allows the same hardware system to precisely adapt to the assistive needs of patients at different stages of rehabilitation.

[0034] The specific working principles of the four rehabilitation intervention models are detailed below: (1) Passive gait pattern When patients are in the early stages of rehabilitation, with weak independent walking ability and obvious gait abnormalities, the system activates a passive gait mode. For example... Figure 4 As shown, in this mode, the lower limb rehabilitation exoskeleton adopts a precise position control mode.

[0035] The specific working process is as follows: Before training begins, the main controller pre-stores the standard gait trajectory set by the clinician. During training, high-precision encoders installed on the hip, knee, and ankle joint modules collect the actual position data of each joint in real time. The main controller performs a difference calculation between the actual collected position data and the preset target gait trajectory to calculate the trajectory deviation value. Subsequently, the main controller, through a PID controller, quickly calculates and outputs a driving torque command based on this deviation value, controlling the motors of each joint to rotate forward or backward, driving the exoskeleton to move the patient's lower limbs to complete periodic walking movements strictly according to the preset gait trajectory. Thus, even when the patient's active muscle contraction is insufficient, the correct gait sensation is input through repeated passive movement, correcting abnormal gait patterns.

[0036] (2) Assistance Mode When the patient's abnormal gait is initially corrected and they have some ability to walk independently but still have insufficient torque in some joints, the system switches to assistive mode. For example... Figure 5 As shown, in this mode, the exoskeleton uses an admittance control mode to encourage patients to actively exert force and achieve compliant assistance.

[0037] The specific working process is as follows: Torque sensors built into the hip, knee, and ankle joints of the exoskeleton collect human-machine interaction torques in real time. Due to the exoskeleton's own weight, joint friction, and inertia, the main controller first preprocesses the collected interaction torque data using gravity compensation algorithms, friction compensation algorithms, and inertia compensation models to accurately offset the aforementioned additional loads and extract the human-machine interaction torque that purely reflects the patient's active force exertion intention. The preprocessed interaction torque is input to the admittance controller, which calculates the gait trajectory correction amount corresponding to the current patient's force exertion state in real time according to pre-set virtual stiffness coefficients, damping coefficients, and inertia coefficients. The system superimposes the correction amount onto the standard target trajectory, dynamically generating a personalized assisted gait trajectory that conforms to gait norms and matches the patient's real-time force exertion intention. Finally, the main controller drives the actuators of each joint to output the corresponding compliant assist torque through motor servo closed-loop control. In this mode, the exoskeleton does not use rigid forced drive. The assistance force is adaptively adjusted according to the degree of active force exerted by the patient. The more active the patient is with assistance, the less additional assistance the exoskeleton provides, and vice versa. This maximizes the stimulation of the patient's residual motor function and guides them toward a standard gait.

[0038] (3) Active mode When the patient's recovery progresses to the middle or late stages and they have developed independent and stable walking ability, the system switches to active mode. For example... Figure 6 As shown, in this mode, the exoskeleton uses a zero-torque follow control mode, acting only as a gait data acquisition device to follow the patient's movement, without providing active assistance or resistance.

[0039] The specific working process is as follows: The main controller receives motion data collected in real time from the encoders and inertial measurement units of each joint. It uses a gravity compensation algorithm to offset the exoskeleton's own weight, a joint friction loss compensation algorithm to offset the frictional resistance of the transmission mechanism, and an inertia compensation model to compensate for the hindering effect caused by motion inertia. The main controller adjusts the output torque of the drive motors in real time to keep it close to the zero torque reference. When the sensors detect that the patient's limb begins to move and generates a small interactive force, the controller responds quickly, driving the corresponding joint motor to rotate synchronously along the patient's movement direction, achieving smooth, imperceptible zero-force following. Throughout the training process, the exoskeleton strictly follows the patient's autonomous movement trajectory and gait rhythm, neither "pulling" nor "dragging" the patient. Simultaneously, the system continuously monitors safety indicators such as gait symmetry, joint range of motion, and interactive force fluctuations. Once gait imbalance or joint angles exceeding safety thresholds are detected, the system immediately triggers a flexible braking protection mechanism, stopping active following and outputting a slight supporting force to prevent falls, ensuring training safety.

[0040] (4) Resistance mode When a patient's rehabilitation enters the final stage, and walking function has largely recovered, requiring further strengthening of lower limb muscle strength and endurance, the system switches to resistance mode. For example... Figure 7 As shown, in this mode, the exoskeleton uses an impedance control mode to provide adjustable reverse resistance for the patient's active walking.

[0041] Taking the hip joint assistive walking exoskeleton as an example, the specific working process is as follows: After the exoskeleton is put on, the high-precision encoder of the hip joint and the inertial measurement unit placed on the thigh bar are activated to collect the angle, angular velocity and gait cycle phase data of the hip joint during the patient's active walking process in real time at a frequency of not less than 100Hz.

[0042] Desired gait trajectory of the hip joint under resistance mode The therapist pre-plans and provides a set of 100 sequence points, fitted into a complete cycle, namely: ;

[0043] Actual feedback angle of hip exoskeleton The data is collected by the encoder at the motor end and calculated using the reduction ratio. ;

[0044] in, i This refers to the reduction ratio of the joint module. The encoder acquires angle values.

[0045] Hip exoskeleton feedback angular velocity Data was collected by an inertial measurement unit (IMU) mounted on the thigh bar. ; in, The angular velocity is acquired by the inertial measurement unit (IMU).

[0046] To ensure that the angle and angular velocity information acquired by the exoskeleton are synchronized with the desired angle information in time, the real-time gait data needs to be segmented into integer cycles and mapped to the corresponding gait phase percentage. Gait phase The calculation is as follows: ; in, This refers to the angle information acquired by the inertial measurement unit (IMU). The properties of the arctangent function indicate that... Normalize it to k The range can then be obtained as follows: ; From this, we can inversely derive the expression for the gait phase with respect to the sequence: ; By utilizing the relationship between gait phase and the feedback angle and angular velocity of the hip joint exoskeleton, 100 sequence points of hip joint feedback angle and angular velocity data can be obtained within each gait cycle: ; ; After the above processing, we can obtain k Desired angle of sequence alignment Sampling angle and the acquisition angular velocity Through such Figure 6 The impedance control mode shown realizes the resistance mode of the hip joint walking exoskeleton. The main controller receives the motion parameters obtained by the above sensors, substitutes them into the preset second-order impedance model (damping-elastic model) for rapid calculation, and generates the target resistance torque for the corresponding joint according to the principle of "reverse motion direction and adaptive intensity". Through the closed-loop control of the drive motor current, the resistance torque is accurately output to the actuators of each joint, providing continuous and adjustable reverse resistance for the patient's walking.

[0047] Specific impedance control implementation examples are as follows: First, calculate the angle and angular velocity feedback error of the hip exoskeleton: ; ; The impedance controller takes angle error and angular velocity error as inputs and plans the output interaction torque: ; in, To output control torque, i.e., resistance torque in this mode, K and B These are the virtual stiffness and damping terms of the impedance controller. In this mode, the larger the values ​​of these two coefficients, the greater the resistance value during resistance training. Specifically, the virtual stiffness coefficient reflects the amplitude of the patient's resistance to the desired gait trajectory while wearing the assistive exoskeleton; a larger amplitude of resistance results in greater resistance. The virtual damping coefficient reflects the effect of the patient's resistance to walking speed while wearing the assistive exoskeleton; a higher walking speed results in greater resistance. These two parameters perfectly satisfy the resistance logic in the resistance training mode.

[0048] To ensure the safety of resistance training, a resistance safety threshold is set. When the patient moves too fast or the joint exceeds its normal range of motion, it causes The system automatically reduces resistance until it cuts off resistance output, triggering a safety protection mechanism. During training, the system simultaneously collects data on the patient's muscle strength output and gait symmetry. Combined with a dynamic closed-loop mechanism of simultaneous intervention and assessment, it judges the patient's training status and tolerance in real time. If the patient's muscle strength improves significantly, the damping coefficient is automatically increased slightly to gradually increase the training intensity. If the patient experiences gait imbalance or abnormal force exertion, the resistance intensity is immediately reduced to ensure that the training intensity always matches the patient's real-time tolerance, achieving personalized and dynamic resistance rehabilitation training.

[0049] Step S4: Online assessment and program updates during the rehabilitation process During the patient's exoskeleton-assisted rehabilitation training following step S3, the system simultaneously performs online assessments and updates the treatment plan. For example... Figure 8 As shown, the overall scheme of this invention constructs a dynamic closed loop of "scale data input → prescription model decision → multi-modal intervention → online evaluation and update".

[0050] Specifically, the exoskeleton's built-in joint encoders, torque sensors, and inertial measurement units continuously collect gait-related data from the patient in real time during training, including but not limited to joint angles, angular velocities, gait cycle phase, stride length, cadence, bilateral symmetry indices, and human-machine interaction torques. This data is transmitted in real time and input into a pre-trained rehabilitation assessment model for stroke patients.

[0051] The training process of this rehabilitation ability assessment model is similar to that of the rehabilitation prescription model in step S2: First, a dataset is constructed consisting of gait data collected by the exoskeleton, training effect scores, and updated labels of rehabilitation intervention patterns manually annotated by physicians; then, the assessment model is trained using machine learning algorithms. This model can analyze the patient's current rehabilitation progress and functional changes based on real-time gait data and output a matching updated exoskeleton rehabilitation intervention plan.

[0052] Step S5: Determine and switch between scheme update conditions After generating the update plan, the system does not switch immediately. Instead, it determines whether the update conditions are met based on preset logic to avoid frequent invalid switches caused by occasional data fluctuations. For example... Figure 9 As shown, the solution update must simultaneously meet the following three sub-conditions: Sub-condition 1: The updated exoskeleton rehabilitation intervention plan differs from the currently implemented plan. If they are the same, no update is needed.

[0053] Sub-condition 2: The update scheme generated in n consecutive iterations remains the same and unchanged. Here, n is a preset positive integer, which can be set by the rehabilitation physician based on clinical experience (e.g., n=3). This condition ensures that the change in rehabilitation ability is a stable trend rather than a random disturbance.

[0054] Sub-condition 3: The update judgment is executed once after each complete rehabilitation training session (e.g., a standard 30-minute training course).

[0055] When all three sub-conditions mentioned above are met simultaneously, the system determines that the patient's rehabilitation ability has undergone a stable, phased change, and the current plan is no longer suitable. At this time, the system sends an update prompt to the physician through the human-computer interaction interface, or automatically replaces the current plan with the updated exoskeleton rehabilitation intervention plan according to preset permissions, and continues to implement the next stage of rehabilitation training for the patient according to the new plan.

[0056] By cyclically executing the above steps S1 to S5, this invention achieves full-cycle, personalized, and dynamically adaptive exoskeleton-assisted rehabilitation for stroke patients from admission assessment to the end of rehabilitation, significantly improving the accuracy, efficiency, and safety of rehabilitation training.

[0057] Example 2: A Multimodal Exoskeleton Rehabilitation System This embodiment provides a multimodal exoskeleton rehabilitation system. The system includes the following functional modules: Data acquisition module: Configured to acquire patient clinical assessment scale data, including Fugue-Meyer rating scale data, Berg balance scale data, and Barthel Index data.

[0058] Prescription generation module: Embedded with a pre-trained stroke rehabilitation prescription model (attention-enhanced random forest model), configured to receive scale data and output an initial exoskeleton rehabilitation intervention plan.

[0059] The modular exoskeleton execution module includes detachable and assembleable lower limb rehabilitation exoskeletons and portable walking assistive exoskeletons, as well as four control algorithm sub-modules integrated into its main controller: passive gait, assisted gait, active gait, and resistive gait. This module is configured to drive the corresponding hardware to execute the selected mode based on the output instructions from the prescription generation module.

[0060] Online assessment and update module: It has an embedded pre-trained rehabilitation ability assessment model, which is configured to receive real-time gait data collected by the exoskeleton's built-in sensors and generate an updated rehabilitation intervention plan.

[0061] Solution switching module: configured to execute the update condition judgment logic described in step S5 of embodiment 1, and control the modular exoskeleton execution module to switch to the new solution when the condition is met.

[0062] Example 3: A computer-readable storage medium Those skilled in the art will understand that Figure 10 The structure shown does not constitute a limitation on the device for generating deep learning-based health management solutions and may include more or fewer components than illustrated.

[0063] The health management solution generation device based on deep learning provided in this application includes a memory 30 and a processor 31. When the processor 31 executes the program stored in the memory 30, it can realize the health management solution generation method based on deep learning.

[0064] Finally, this application also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps described in the above method embodiments.

[0065] It is understood that if the methods in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and executes all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0066] The foregoing provides a detailed description of a method, apparatus, device, and medium for generating a health management solution based on deep learning, as provided in this application. The various embodiments in the specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0067] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A multimodal exoskeleton rehabilitation method, characterized in that, Includes the following steps: Step S1: Obtain patient clinical assessment scale data, including Fugue-Meier rating scale data, Berg balance scale data, and Barthel Index data; Step S2: Input the scale data into the pre-trained stroke rehabilitation prescription model, and the rehabilitation prescription model outputs an initial exoskeleton rehabilitation intervention plan that matches the patient's current rehabilitation ability; wherein, the rehabilitation prescription model is trained based on a high-quality rehabilitation dataset, and the high-quality rehabilitation dataset consists of the scale data and rehabilitation intervention mode labels manually annotated by physicians; Step S3: Based on the initial exoskeleton rehabilitation intervention plan, select the corresponding exoskeleton type and rehabilitation intervention mode from the modular exoskeleton system, and implement rehabilitation training for the patient; wherein, the exoskeleton type includes lower limb rehabilitation exoskeleton suitable for the early and middle stages of rehabilitation, and portable walking assistive exoskeleton suitable for the middle and late stages of rehabilitation; the rehabilitation intervention mode includes passive gait mode, assisted mode, active mode and resistance mode. Step S4: During rehabilitation training, gait-related data of the patient is collected in real time through the built-in sensors of the exoskeleton, and the gait-related data is input into a pre-trained rehabilitation ability assessment model to generate an updated exoskeleton rehabilitation intervention plan; wherein, the rehabilitation ability assessment model is trained based on the preset exoskeleton data collection dataset and the rehabilitation intervention mode updated labels manually labeled by the physician. Step S5: Determine whether the conditions for updating the treatment plan are met. If they are met, replace the current treatment plan with the updated exoskeleton rehabilitation intervention plan and continue to provide rehabilitation training to the patient.

2. The multimodal exoskeleton rehabilitation method according to claim 1, characterized in that, The stroke rehabilitation prescription model in step S2 adopts an attention-enhanced random forest model. The attention-enhanced random forest model assigns differentiated weights to each assessment dimension in the Fugue-Meier rating scale data, the Berg balance scale data, and the Barthel Index data through an attention mechanism to capture the correlation weights between different scale characteristics and rehabilitation intervention programs.

3. The multimodal exoskeleton rehabilitation method according to claim 1, characterized in that, The passive gait mode in step S3 specifically involves using a position control mode on the lower limb rehabilitation exoskeleton to drive the exoskeleton joints to move the patient's limbs strictly according to a preset standard gait trajectory in order to correct abnormal gait.

4. The multimodal exoskeleton rehabilitation method according to claim 1, characterized in that, The assist mode in step S3 is as follows: the torque sensor built into the exoskeleton joint collects the human-machine interaction torque in real time, and after gravity compensation, it is input into the admittance controller. The admittance controller dynamically and flexibly corrects the standard gait trajectory according to the preset virtual stiffness coefficient, damping coefficient and inertia coefficient, and generates a personalized assist gait trajectory that fits the patient's real-time force exertion state.

5. The multimodal exoskeleton rehabilitation method according to claim 1, characterized in that, The active mode in step S3 is as follows: the exoskeleton adopts a zero-torque following control mode. Through gravity compensation, joint friction loss compensation and inertia compensation, the output torque of the drive motor is adjusted to approach zero, and it follows the patient's autonomous movement trajectory to achieve imperceptible zero-force following. At the same time, gait data is monitored in real time and safety protection is triggered.

6. The multimodal exoskeleton rehabilitation method according to claim 1, characterized in that, The resistance mode in step S3 is as follows: the exoskeleton adopts an impedance control mode to establish a dynamic relationship between joint torque and position and velocity, and provides adjustable damping resistance opposite to the direction of movement during the patient's active movement; the impedance control mode is based on the error between the feedback angle and the desired angle of the hip joint exoskeleton, as well as the error between the feedback angular velocity and the desired angular velocity, and calculates the output control torque as the resistance torque through virtual stiffness term and virtual damping term.

7. The multimodal exoskeleton rehabilitation method according to claim 1, characterized in that, In step S4, the exoskeleton's built-in sensors include joint end encoders, torque sensors, and inertial measurement units. The collected gait-related data include joint angles, angular velocities, gait cycle phases, and human-machine interaction torques.

8. The multimodal exoskeleton rehabilitation method according to claim 1, characterized in that, In step S5, determining whether the scheme update conditions are met requires that the following sub-conditions be met simultaneously: Sub-condition 1: The updated exoskeleton rehabilitation intervention program differs from the current program; Sub-condition 2: The update scheme generated in n consecutive times remains unchanged, where n is a preset positive integer; Sub-condition 3: An updated plan is generated for each complete rehabilitation training session.

9. A multimodal exoskeleton rehabilitation system, characterized in that, include: The data acquisition module is used to acquire patient clinical assessment scale data, including Fugue-Meier rating scale data, Berg balance scale data, and Barthel Index data. The prescription generation module is used to input the scale data into a pre-trained stroke rehabilitation prescription model and output an initial exoskeleton rehabilitation intervention plan that matches the patient's current rehabilitation ability. The modular exoskeleton execution module is used to select the exoskeleton type and rehabilitation intervention mode according to the initial exoskeleton rehabilitation intervention plan to carry out rehabilitation training for the patient. The exoskeleton type includes a lower limb rehabilitation exoskeleton and a portable walking assistive exoskeleton. The rehabilitation intervention mode includes a passive gait mode, an assistive mode, an active mode, and an resistance mode. The online assessment and update module is used to collect gait data through the exoskeleton's built-in sensors during training and input it into the rehabilitation ability assessment model to generate an updated exoskeleton rehabilitation intervention plan. The scheme switching module is used to replace the current scheme with the updated scheme when preset update conditions are met.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multimodal exoskeleton rehabilitation method as described in any one of claims 1 to 8.

Images