Method and system for determining a training regimen for home-based conservative treatment of patients with knee motion impairment

Abstract

This invention discloses a method and system for determining training programs for home-based conservative treatment of patients with knee joint sports injuries. By acquiring the patient's multimodal data, an observable state vector is constructed, the patient's recovery stage, a set of injury classification diagnoses, and a set of injury constraints corresponding to the recovery stage are determined to obtain the training program for round t. Based on the feedback information after the patient executes the training program for round t, it is determined whether the training program switching conditions are met. If the switching conditions are met, the feedback information and the historical context of the training program for round t are input into the program generation strategy model to obtain the training program for round t+1. This method can improve the individualization, safety, and treatment course adaptability of home-based conservative treatment prescriptions.

Description

Technical Field

[0001] This invention relates to the fields of sports medicine and rehabilitation medicine, and in particular to a method and system for determining a training program for conservative home treatment of patients with knee joint sports injuries. Background Technology

[0002] The core of conservative treatment for knee sports injuries lies in gradually improving joint range of motion, strength, neuromuscular control, and overall function, while ensuring tissue healing and symptom control, ultimately restoring daily or athletic abilities. Current techniques typically have the following limitations:

[0003] Static prescriptions: Many home training programs are template-based and fixed processes, lacking a closed-loop mechanism for "dynamic parameter adjustment as recovery progresses." Difficulty in standardizing treatments for multiple conditions / complications: Different injury types (such as ligament and meniscus coexistence, patellofemoral pain with synovitis, etc.) have different contraindications and progression logics, making it difficult for manual rules to cover all combinations. Lack of verifiable iterative error correction mechanisms: If problems such as pain rebound, increased swelling, or compensatory movements occur after prescription generation, the system usually cannot effectively "locate errors-correct-re-verify" through multiple rounds of interaction.

[0004] Therefore, there is a need for a technical solution that can utilize wearable sensing and intelligent agent assessment capabilities to construct a closed loop of "assessment-prescription-training-feedback-reassessment-represcription" under home-based conservative treatment conditions, while taking into account safety constraints, individual differences, stage progression, and computing power scheduling efficiency. Summary of the Invention

[0005] In view of the above, the present invention aims to provide a method and system for determining a training program for home conservative treatment of patients with knee joint sports injuries, so as to solve the aforementioned technical problems.

[0006] The technical solution adopted in this invention is as follows:

[0007] This invention provides a method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries, including:

[0008] In the t-th round of training, multimodal data of the patient is acquired, including at least language consultation data, machine vision physical examination data, motion sensing training data, and medical imaging data.

[0009] Based on the multimodal data, an observable state vector of the patient is constructed;

[0010] Based on the observable state vector, the patient's recovery stage and the set of damage classification diagnoses are determined;

[0011] Based on the recovery stage and the damage classification diagnosis set, the damage constraint set corresponding to the recovery stage is obtained;

[0012] The observable state vector, the set of damage constraints, and the historical context of the training scheme in round t are input into the scheme generation strategy model to obtain the training scheme in round t.

[0013] Obtain feedback information from the patient after executing the training program in round t;

[0014] Based on the feedback information and the recovery phase, determine whether the training scheme switching conditions are met;

[0015] If the switching condition is met, the feedback information and the historical context of the training scheme in round t are input into the scheme generation strategy model to obtain the training scheme in round t+1.

[0016] Optionally, based on the multimodal data, an observable state vector of the patient is constructed, including:

[0017] Extract consultation features from the language consultation data;

[0018] Extract visual physical examination features from the machine vision physical examination data;

[0019] Extract training sensing features from the motion sensing training data;

[0020] Extract image analysis features from the medical image data;

[0021] Obtain patient compliance characteristics and individual profile characteristics;

[0022] The aforementioned consultation features, visual examination features, training sensing features, image analysis features, compliance features, and individual profile features are fused to obtain an observable state vector.

[0023] Optionally, the injury classification diagnosis set can be obtained based on the medical history features, the visual examination features, and the image analysis features.

[0024] Optionally, based on the recovery stage and the damage classification diagnostic set, a damage constraint set corresponding to the recovery stage is obtained, including:

[0025] Based on the recovery stage and the damage classification diagnosis set, a diagnosis constraint set, a stage constraint set, and a risk constraint set are obtained;

[0026] The set of diagnostic constraints, the set of stage constraints, and the set of risk constraints are merged to obtain the set of damage constraints.

[0027] Optionally, the observable state vector, the set of damage constraints, and the historical context of the training scheme in round t are input into the scheme generation policy model to obtain the training scheme in round t, including:

[0028] ;

[0029] ;

[0030] in, For the training scheme in round t, For the historical context of the training scheme in round t, Generate a strategy model for the solution. For the k-th training action or task in round t, For the action parameter vector, This represents the number of actions in this round.

[0031] Optionally, the historical context of the training scheme in the t-th round is:

[0032] ;

[0033] in, For the historical context of the training scheme in round t, For historical training programs, For historical feedback information, The length of the sliding memory window and ≥ 1.

[0034] Optionally, based on the feedback information and the recovery phase, it is determined whether the training scheme switching conditions are met, including:

[0035] Based on the feedback information, a recovery score and a risk score are obtained;

[0036] When both conditions are met , , When the switching conditions are met;

[0037] Otherwise, the switching conditions are not met;

[0038] in, For the recovery score in round t, For the recovery score in round t-1, To score the risk, This represents the minimum expected increase threshold corresponding to the recovery phase. This represents the maximum permissible risk threshold corresponding to the recovery phase. This is a risk warning message. .

[0039] Optionally, the feedback information and the historical context of the training scheme in round t are input into the scheme generation policy model to obtain the training scheme in round t+1, including:

[0040] ;

[0041] in, For the training scheme in round t+1, This is the feedback information for the t-th round.

[0042] Optionally, the training process of the scheme generation strategy model is as follows:

[0043] A multi-round training scheme trajectory dataset was collected under the output distribution of the scheme generation strategy model;

[0044] Based on the training scheme trajectory dataset, an improved training scheme is constructed through expert distillation or self-distillation.

[0045] The scheme generation strategy model is updated using a reward-weighted supervised objective function. This invention also provides a training scheme determination system for home-based conservative treatment of patients with knee joint sports injuries, comprising:

[0046] The data acquisition module is used to acquire multimodal data from patients;

[0047] The feature extraction and state construction module is used to construct an observable state vector of the patient based on the multimodal data;

[0048] The staging and constraint generation module is used to determine the patient's recovery stage and damage classification diagnosis set based on the observable state vector; and to obtain the damage constraint set corresponding to the recovery stage based on the recovery stage and the damage classification diagnosis set.

[0049] The scheme generation and adjustment module is used to input the observable state vector, the set of damage constraints, and the historical context of the training scheme in round t into the scheme generation strategy model to obtain the training scheme in round t.

[0050] The training feedback and scoring module is used to obtain feedback information from the patient after executing the training program in round t; and to determine whether the conditions for switching training programs are met based on the feedback information and the recovery stage.

[0051] The scheme switching and output module is used to input feedback information and the historical context of the training scheme in round t into the scheme generation strategy model to obtain the training scheme in round t+1. The above-mentioned scheme of the present invention includes at least the following beneficial effects:

[0052] The above-described solution of the present invention, in the t-th training round, acquires multimodal data of the patient, including at least language consultation data, machine vision physical examination data, motion sensing training data, and medical imaging data; constructs an observable state vector of the patient based on the multimodal data; determines the patient's recovery stage and damage classification diagnostic set based on the observable state vector; obtains a damage constraint set corresponding to the recovery stage based on the recovery stage and the damage classification diagnostic set; inputs the observable state vector, the damage constraint set, and the historical context of the t-th training scheme into the scheme generation strategy model to obtain the t-th training scheme; acquires feedback information from the patient after executing the t-th training scheme; determines whether the training scheme switching condition is met based on the feedback information and the recovery stage; if the switching condition is met, inputs the feedback information and the historical context of the t-th training scheme into the scheme generation strategy model to obtain the (t+1)-th training scheme. This can improve the individualization, safety, and treatment course adaptability of home-based conservative treatment prescriptions. Attached Figure Description

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings, wherein:

[0054] Figure 1 This is a schematic diagram of a method for determining a training program for conservative home treatment of patients with knee joint sports injuries, as provided in an embodiment of the present invention. Detailed Implementation

[0055] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0056] This invention proposes an embodiment of a method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries. Specifically, as follows: Figure 1 As shown, it includes:

[0057] Step 11: In the t-th round of training, acquire the patient's multimodal data, which includes at least language consultation data, machine vision physical examination data, motion sensing training data, and medical imaging data.

[0058] Step 12: Based on the multimodal data, construct the patient's observable state vector;

[0059] Step 13: Determine the patient's recovery stage and damage classification diagnostic set based on the observable state vector;

[0060] Step 14: Based on the recovery stage and the damage classification diagnosis set, obtain the damage constraint set corresponding to the recovery stage;

[0061] Step 15: Input the observable state vector, the set of damage constraints, and the historical context of the training scheme in round t into the scheme generation strategy model to obtain the training scheme in round t.

[0062] Step 16: Obtain feedback information from the patient after executing the training program in round t;

[0063] Step 17: Based on the feedback information and the recovery phase, determine whether the training scheme switching conditions are met;

[0064] Step 18: If the switching condition is met, input the feedback information and the historical context of the training scheme in round t into the scheme generation strategy model to obtain the training scheme in round t+1.

[0065] In this embodiment, in step 11, data such as knee joint acceleration, angular velocity, pressure / plantar force, electromyography, and joint angle estimation are collected through wearable devices or external sensors; video of the patient's movements is collected through a mobile phone or camera; medical imaging data is obtained through MRI / CT / ultrasound / DR images or reports; and language consultation data is obtained through text / speech transcription and scales.

[0066] In step 12, based on the multimodal data, an observable state vector of the patient is constructed:

[0067] The system extracts consultation features from the language consultation data; visual examination features from the machine vision examination data; training sensing features from the motion sensing training data; and image analysis features from the medical image data. Simultaneously, it acquires patient compliance features and individual profile features. The consultation features, visual examination features, training sensing features, image analysis features, compliance features, and individual profile features are then fused to obtain an observable state vector.

[0068] At the t-th evaluation / training loop closure time, the observable state vector expression composed of concatenated multimodal features is as follows:

[0069] ;

[0070] in, The observable state vector;

[0071] For characteristics of the consultation, such as pain VAS, morning stiffness, swelling complaints, clicking / popping sounds, post-training response, etc.;

[0072] Visual physical examination features, such as knee valgus angle, hip-knee-ankle alignment, trunk lateral tilt, and gait symmetry;

[0073] For image analysis features, such as the degree of anterior cruciate ligament signal abnormality, the location and shape of meniscus tear, synovial thickening, etc.;

[0074] To train sensor characteristics, such as flexion-extension angle range, peak angular velocity, left-right symmetry, motion stability, repetition efficiency, load / impact index, etc.

[0075] Compliance characteristics, such as completion rate, missed practice rate, RPE subjective effort, sleep, and other influencing factors;

[0076] Individual profile characteristics, such as age, BMI, exercise goals, medical history, and equipment conditions.

[0077] In step 13, the patient's recovery stage is first determined based on the observable state vector. The recovery stage includes the patient's current inflammation control, ROM / control recovery, strength endurance, and functional / regression movement.

[0078] ;

[0079] For example, This indicates the acute inflammation control phase. Indicates activity level / control recovery period. This indicates the period of strength and endurance rebuilding. This indicates the functional / regression phase, etc.

[0080] In step 14, based on the observable state vector, the recovery stage, and the damage classification diagnosis set, the damage constraint set corresponding to the recovery stage is obtained:

[0081] ;

[0082] in, This is a diagnostic set for injury classification, obtained by fusing and judging characteristics from medical history, visual examination, and image analysis. The diagnostic set for injury classification includes ligament / meniscus / patellofemoral pain / synovitis / tendinitis / contusion, etc., and multiple conditions may coexist.

[0083] The damage constraint set is the union of damage taboos, phase rules, and risk triggering rules, where... For injury-specific contraindications and protective restrictions (such as limiting deep knee flexion at certain stages, avoiding rotational shear loads, etc.); These are the phase goals and advanced rules; For example: risk-triggered constraints (pain rebound, increased swelling, deterioration of movement quality, etc.).

[0084] In step 15, the observable state vector, the set of damage constraints, and the historical context of the training scheme in round t are input into the scheme generation policy model to obtain the training scheme in round t:

[0085] ;

[0086] ;

[0087] in, For the training scheme in round t, For the historical context of the training scheme in round t, Generate a strategy model for the solution. For the k-th training action or task in round t, For the action parameter vector, The number of moves in this round, For natural language interpretation.

[0088] Here, during the solution generation process, the solution generation strategy model will initially generate N candidate solutions. Constraint filtering is performed using the damage constraint set to eliminate candidates that do not meet the constraints. The remaining candidates are then subjected to "action set similarity clustering + parameter median / weighted average" to output the final training scheme. And through Simultaneously output explanatory text: objectives, precautions, risk warnings, and reasons for adjustment.

[0089] In step 16, after the t-th round of training, obtain the patient's feedback information after implementing the t-th round of training plan. Feedback information is automatically generated by the system, including: during training: real-time / near real-time monitoring of movement quality (video / sensor); after training: collection of pain / swelling changes, next-day response, RPE, completion rate, etc.; risk warnings, abnormal data fluctuations, etc.; error correction instructions: such as "End-effector ROM load is too high / inward twitching is increased, please adjust the movement or reduce the range / sets / rhythm".

[0090] Step 17: Based on the feedback information and the recovery phase, determine whether the training scheme switching conditions are met.

[0091] Based on the feedback information, a recovery score is obtained. and risk score ;

[0092] When both conditions are met , , When the switching conditions are met;

[0093] Otherwise, the switching conditions are not met;

[0094] in, For the recovery score in round t, For the recovery score in round t-1, To score the risk, This represents the minimum expected increase threshold corresponding to the recovery phase. This represents the maximum permissible risk threshold corresponding to the recovery phase. This is a risk warning message. .

[0095] Recovery rating:

[0096] ;

[0097] ;

[0098] ;

[0099] in, The k-th normalized recovery index (ROM, strength symmetry, function scale, movement quality, endurance, etc.).

[0100] Risk Score:

[0101] ;

[0102] ;

[0103] ;

[0104] in, For the qth risk indicator (pain rebound, increased swelling, movement collapse, peak impact, suspected instability, etc.).

[0105] Binary success indication:

[0106] .

[0107] In step 18, when When the switching condition is met, the training scheme for round t+1 is obtained from the feedback information and the historical context of the training scheme in round t, input into the scheme generation strategy model.

[0108] ;

[0109] in, For the historical context of the training scheme in round t:

[0110] ;

[0111] For historical training programs, For historical feedback information, The length of the sliding memory window and ≥1.

[0112] The training process of the scheme generation strategy model in this embodiment is as follows:

[0113] Data sets of trajectory schemes from multiple rounds of training were collected under the output distribution of the scheme generation strategy model π. Trajectory data includes observable state vectors Historical context Training program Feedback information Recovery score and risk score ;

[0114] ;

[0115] Where i is the patient index and t is the iteration index within a round. For historical context state, For the current training scheme, This is a binary success indicator for the current training scheme.

[0116] Method 1: Construct an improved generation strategy model through expert distillation or self-distillation.

[0117] Method 1: Output the training scheme given the historical context through expert distillation:

[0118] ;

[0119] in, For expert strategy models, This is a privileged status, such as complete image segmentation, details of offline physical examination, doctor's conclusions, and long-term follow-up outcomes.

[0120] Method 2: Training is performed through self-distillation. Multiple candidate prescriptions are sampled from the current policy for the same state, and the one with the highest score is selected as the improvement plan.

[0121] , ;

[0122] And use a reward predictor / security evaluator to select the best option:

[0123] ;

[0124] For reward predictors or security evaluators.

[0125] Furthermore, a reward-weighted supervised update strategy is adopted, enabling the model to learn high-quality corrections while also utilizing some effective information from imperfect trajectories.

[0126] Reward-weighted objectives:

[0127] ;

[0128] ;

[0129] in, The training set includes the improved scheme; This is a temperature coefficient used to control the weight difference between high and low reward samples. ; The average of multiple reward attempts for the same patient / same stage / same problem.

[0130] Furthermore, during home deployment, the system cannot reliably obtain offline physical examinations and complete image details. Therefore, if experts rely too heavily on privileged information, student strategies will be difficult to reproduce during deployment, resulting in poor feasibility. This embodiment addresses this issue through a "constrained privileged expert" mechanism.

[0131] Specifically, the training scheme output is:

[0132] ;

[0133] Interactive data aggregation and updates are performed on the solution generation strategy model:

[0134] ;

[0135] in, This is the supervised loss function.

[0136] Furthermore, this embodiment also includes multi-task scheduling under heterogeneous computing power, including language consultation, visual physical examination, image analysis, prescription generation, strategy optimization / update, etc. To reduce end-to-end latency and cost, this workflow is jointly optimized and scheduled on a heterogeneous device graph.

[0137] Specifically, when deployed at home, the modules for language consultation, machine vision physical examination, automatic image analysis, prescription generation, and consistency selection are constructed as a task computation graph G=(V,E). On a heterogeneous device topology graph D=(U,L) encompassing mobile devices, edge devices, and the cloud, a partitioning strategy is determined through joint optimization. With assignment strategy To minimize end-to-end latency or computational cost, and to satisfy resource constraints including memory constraints and bandwidth constraints; the end-to-end latency cost aggregation model considers the task parallelism coefficient. ,in =0 indicates approximately serial execution. =1 indicates near-complete parallel execution.

[0138] Joint optimization objective function: ;

[0139] in, For partitioning / parallelization strategies (task slices), the number of task slices shall not exceed the number of available devices; To assign task segments to specific devices, This is an estimation function for end-to-end delay or unit time cost.

[0140] Cost aggregation model (parallelism parameterization):

[0141] ;

[0142] in, This is the task parallelism coefficient (0 indicates approximately serial, 1 indicates approximately fully parallel). .

[0143] Specific Example 1: Home conservative treatment for patellofemoral pain (PFPS) complicated with mild synovitis:

[0144] Initial assessment: Verbal history revealed pain when going up and down stairs and pain when standing up after sitting for a long time; visual examination identified knee valgus during single-leg squatting; sensor data showed increased pain at the end of the squat angle range; imaging / reports indicated mild inflammatory changes in the synovium.

[0145] Phase determination: (Inflammation control + motor control recovery).

[0146] Prescription generation: Select hip abduction / external rotation strength, quadriceps isometric, closed-chain small-angle control, gait and step control from the action library; parameter limits deep flexion angle and high impact.

[0147] Training feedback: If and Triggering constraints The system generates error correction feedback information. The report indicates "excessive end-angle load / increased tuck in motion," and suggests reducing ROM or adjusting the motion in the next round, as well as increasing hip control pre-activation.

[0148] Multiple rounds of introspective adjustments: and The input is then concatenated to the next round of policy input, generating a more secure one. And through majority consistency, it maintains stable output under noisy conditions.

[0149] Specific Example 2: Home Conservative Treatment for Meniscus Injury (Posterior Horn) Combined with Mild Ligament Injury:

[0150] Automatic image analysis outputs the characteristics of the meniscus tear area and severity; visual examination indicates insufficient rotational control; sensory indications show that the peak internal rotational angular velocity is too high when squatting.

[0151] constraint library Add the combined constraint of "avoiding early deep knee flexion + rotational shear load"; stage Primarily based on ROM and control, settings and More conservative.

[0152] The prescription gradually increases the load in multiple iterations: when Meets the standards and When the threshold is below, the system allows entry. It also increases the intensity of strength training while maintaining rotational control training and the quality threshold of movement.

[0153] A specific implementation example 3: Implementation of heterogeneous scheduling in home-cloud collaboration:

[0154] Mobile devices: Real-time sensor processing, lightweight assessment of motion video, prescription display and interaction;

[0155] Cloud-based: Image remodeling analysis, policy training / distillation update, complex consistency evaluation;

[0156] Through scheduling optimization (partitioning) Assignment ) and cost model Control end-to-end latency; when network is poor or budget is limited, it can be downgraded to local policy + lightweight visual assessment, with cloud updates delayed.

[0157] In summary, the method for determining training programs for home-based conservative treatment of knee joint sports injuries in this embodiment integrates multimodal information such as motion sensing training data, verbal consultation, machine vision physical examination, and automatic analysis of medical images. It constructs an observable patient state and determines the recovery stage, generating a structured training program that meets injury-specific contraindications, stage goals, and risk thresholds. In the training-feedback closed loop, historical training programs and feedback are used as contextual inputs, and a multi-round introspective strategy improvement mechanism iteratively corrects the training program. During the training phase, constrained expert feedback with access to privileged clinical information is introduced for distillation, enabling the deployment phase to output high-quality training programs without relying on privileged information. Furthermore, heterogeneous computing power scheduling optimization reduces end-to-end latency and cost of the multi-model workflow.

[0158] Through a multimodal assessment-solution-feedback-reassessment mechanism, the solution parameters are dynamically adjusted as recovery progresses, avoiding "overtraining / undertraining" caused by fixed templates. Incorporating failed solutions and feedback into the status allows for the identification of key factors leading to pain rebound and deterioration in movement quality, enabling targeted corrections of movement selection and parameters in subsequent rounds, reducing the probability of "getting worse with each change." Constrained privileged experts and feasibility controls ensure the model fully absorbs the "error-correction advantage" from complete imaging and doctor's examinations during training, while still outputting high-quality solutions in actual home settings based solely on available data. Risk scoring, risk identification, and constraint sets work together to ensure that solution generation naturally meets injury contraindications and stage protection principles. Once a threshold is triggered, automatic downgrading / pause / recommendation for follow-up visits can be implemented, enhancing home safety boundaries.

[0159] Embodiments of the present invention also provide a system for determining a training program for home conservative treatment of patients with knee joint sports injuries, comprising:

[0160] The data acquisition module is used to acquire multimodal data from patients;

[0161] The feature extraction and state construction module is used to construct an observable state vector of the patient based on the multimodal data;

[0162] The staging and constraint generation module is used to determine the patient's recovery stage and damage classification diagnosis set based on the observable state vector; and to obtain the damage constraint set corresponding to the recovery stage based on the recovery stage and the damage classification diagnosis set.

[0163] The scheme generation and adjustment module is used to input the observable state vector, the set of damage constraints, and the historical context of the training scheme in round t into the scheme generation strategy model to obtain the training scheme in round t.

[0164] The training feedback and scoring module is used to obtain feedback information from the patient after executing the training program in round t; and to determine whether the conditions for switching training programs are met based on the feedback information and the recovery stage.

[0165] The scheme switching and output module is used to input the feedback information and the historical context of the training scheme in round t into the scheme generation strategy model to obtain the training scheme in round t+1.

[0166] It should be noted that this system is the system corresponding to the above method. All implementation methods in the above method embodiments are applicable to the embodiments of this system and can achieve the same technical effect.

[0167] An embodiment of the present invention also provides a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described in the above embodiments. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0168] In this embodiment of the invention, a computer-readable storage medium is also provided, storing instructions that, when executed on a computer, cause the computer to perform the method described in the above embodiments. All implementations of the methods described in the above embodiments are applicable to this embodiment and can achieve the same technical effect.

[0169] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0170] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0171] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0172] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0173] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0174] If the aforementioned functions 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 invention, essentially, or the part that contributes to the prior art, or a portion 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 includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0175] Furthermore, it should be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or a combination thereof. This is something that those skilled in the art can achieve by using their basic programming skills after reading the description of the present invention.

[0176] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing device. The computing device can be a known general-purpose device. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0177] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries, characterized in that, include: In the t-th round of training, multimodal data of the patient is acquired, including at least language consultation data, machine vision physical examination data, motion sensing training data, and medical imaging data. Based on the multimodal data, an observable state vector of the patient is constructed; Based on the observable state vector, the patient's recovery stage and the set of damage classification diagnoses are determined; Based on the recovery stage and the damage classification diagnosis set, the damage constraint set corresponding to the recovery stage is obtained; The observable state vector, the set of damage constraints, and the historical context of the training scheme in round t are input into the scheme generation strategy model to obtain the training scheme in round t. Obtain feedback information from the patient after executing the training program in round t; Based on the feedback information and the recovery phase, determine whether the training scheme switching conditions are met; If the switching condition is met, the feedback information and the historical context of the training scheme in round t are input into the scheme generation strategy model to obtain the training scheme in round t+1.

2. The method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries according to claim 1, characterized in that, Based on the multimodal data, an observable state vector of the patient is constructed, including: Extract consultation features from the language consultation data; Extract visual physical examination features from the machine vision physical examination data; Extract training sensing features from the motion sensing training data; Extract image analysis features from the medical image data; Obtain patient compliance characteristics and individual profile characteristics; The aforementioned consultation features, visual examination features, training sensing features, image analysis features, compliance features, and individual profile features are fused to obtain an observable state vector. ； in, For the characteristics of the consultation, For visual physical examination characteristics, For image analysis features, To train sensor features, As a characteristic of compliance, Characteristics of individual records.

3. The method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries according to claim 2, characterized in that, The injury classification diagnosis set is obtained based on the described questioning characteristics, the described visual examination characteristics, and the described image analysis characteristics.

4. The method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries according to claim 1, characterized in that, Based on the recovery stage and the damage classification diagnosis set, a damage constraint set corresponding to the recovery stage is obtained, including: Based on the recovery stage and the damage classification diagnosis set, a diagnosis constraint set, a stage constraint set, and a risk constraint set are obtained; The set of diagnostic constraints, the set of stage constraints, and the set of risk constraints are merged to obtain the set of damage constraints.

5. The method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries according to claim 1, characterized in that, The observable state vector, the set of damage constraints, and the historical context of the training scheme in round t are input into the scheme generation policy model to obtain the training scheme in round t, including: ； ； in, For the training scheme in round t, For the historical context of the training scheme in round t, Generate a strategy model for the solution. For the k-th training action or task in round t, For the action parameter vector, This represents the number of actions in this round.

6. The method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries according to claim 5, characterized in that, The historical context of the training scheme in round t is: ； in, For the historical context of the training scheme in round t, For historical training programs, For historical feedback information, The length of the sliding memory window and ≥ 1.

7. The method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries according to claim 1, characterized in that, Based on the feedback information and the recovery phase, determine whether the training scheme switching conditions are met, including: Based on the feedback information, a recovery score and a risk score are obtained; When both conditions are met , , When the switching conditions are met; Otherwise, the switching conditions are not met; in, For the recovery score in round t, For the recovery score in round t-1, To score risk, This represents the minimum expected increase threshold corresponding to the recovery phase. This represents the maximum permissible risk threshold corresponding to the recovery phase. This is a risk warning message. .

8. The method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries according to claim 1, characterized in that, By inputting the feedback information and the historical context of the training scheme in round t into the scheme generation policy model, the training scheme in round t+1 is obtained, including: ； in, For the training scheme in round t+1, This is the feedback information for the t-th round.

9. The method for determining a training program for home-based conservative treatment of patients with knee joint sports injuries according to claim 1, characterized in that, The training process of the scheme generation strategy model is as follows: A multi-round training scheme trajectory dataset was collected under the output distribution of the scheme generation strategy model; Based on the training scheme trajectory dataset, an improved training scheme is constructed through expert distillation or self-distillation. The proposed strategy model is updated using a reward-weighted supervised objective function.

10. A system for determining a training program for home-based conservative treatment of patients with knee joint sports injuries, characterized in that, include: The data acquisition module is used to acquire multimodal data from patients; The feature extraction and state construction module is used to construct an observable state vector of the patient based on the multimodal data; The staging and constraint generation module is used to determine the patient's recovery stage and damage classification diagnosis set based on the observable state vector; and to obtain the damage constraint set corresponding to the recovery stage based on the recovery stage and the damage classification diagnosis set. The scheme generation and adjustment module is used to input the observable state vector, the set of damage constraints, and the historical context of the training scheme in round t into the scheme generation strategy model to obtain the training scheme in round t. The training feedback and scoring module is used to obtain feedback information from the patient after executing the training program in the t-th round. Based on the feedback information and the recovery phase, determine whether the training scheme switching conditions are met; The scheme switching and output module is used to input the feedback information and the historical context of the training scheme in round t into the scheme generation strategy model to obtain the training scheme in round t+1.

Images