Orthopedic patient rehabilitation process tracking method and device, storage medium and equipment

Abstract

The invention provides an orthopedic patient rehabilitation process tracking method and device, a storage medium and equipment, and the method comprises the steps: carrying out the feature extraction of multi-modal rehabilitation data, forming a rehabilitation feature set of each rehabilitation process, and carrying out the feature extraction of the rehabilitation feature set in the same rehabilitation process; in different rehabilitation processes, two rehabilitation features are exhaustively extracted from each rehabilitation feature set to form a plurality of different first rehabilitation feature pairs, and in different rehabilitation processes, one rehabilitation feature is respectively exhaustively extracted from each two rehabilitation feature sets according to the associated information to form a plurality of different second rehabilitation feature pairs; constructing a fusion feature set according to the plurality of first and second rehabilitation feature pairs, inputting the fusion feature set into a preset rehabilitation state evaluation model, and outputting and obtaining rehabilitation state evaluation results of the orthopedic patient in different rehabilitation processes, so as to integrate multi-modal data, mine a relationship among different modal data, and obtain rehabilitation state evaluation results of the orthopedic patient in different rehabilitation processes. And the data relationship between the same or different rehabilitation processes is mined, so that the rehabilitation state of the patient in each rehabilitation process can be comprehensively and accurately evaluated.

Description

Technical Field

[0001] The present application relates to the field of orthopedic rehabilitation medical technology, and specifically to a method, device, storage medium and equipment for tracking the rehabilitation progress of orthopedic patients. Background Art

[0002] In orthopedics, total knee replacement (TKA) involves replacing a deformed knee joint caused by osteoarthritis or rheumatoid arthritis with an artificial material. It is an effective surgical procedure for treating end-stage knee osteoarthritis from various causes. While TKA is an effective treatment for severe knee problems, postoperative recovery monitoring is crucial.

[0003] However, when processing multimodal rehabilitation data from different data sources, traditional methods for tracking the postoperative rehabilitation process of orthopedic patients do not adequately mine the relationships between different modal data, nor do they adequately mine the relationships between data of the same or different rehabilitation processes. This makes it impossible to fully reflect the rehabilitation process, resulting in low accuracy in rehabilitation status assessment. For example, in the technical solution with patent application number 202411524542.9, the medical history, clinical information, and physical examination data of orthopedic patients are collected to record the patient’s initial motor ability and functional status as baseline data. Through a preset evaluation cycle, the patient’s rehabilitation progress is monitored in real time, and through the system’s data analysis function, the patient’s rehabilitation effect is quantified, and the stage effectiveness of the rehabilitation plan is evaluated, thereby managing the rehabilitation process more scientifically. However, this technical solution does not fully mine the relationships between different modal data, making it difficult to fully reflect the rehabilitation process. Summary of the Invention

[0004] The main purpose of this application is to provide a method, device, storage medium and equipment for tracking the rehabilitation progress of orthopedic patients, so as to realize the extraction and fusion of different modal features, fully explore the relationship between different modal data, and explore the data relationship between the same or different rehabilitation processes, so as to fully reflect the rehabilitation progress of orthopedic patients and improve the accuracy of rehabilitation status assessment.

[0005] In order to achieve the above-mentioned purpose of the invention, the present application provides a method for tracking the rehabilitation progress of orthopedic patients, comprising:

[0006] To obtain multimodal rehabilitation data of orthopedic patients at different stages of rehabilitation after total knee replacement;

[0007] performing feature extraction on the multimodal rehabilitation data to form a rehabilitation feature set for each rehabilitation process, wherein each rehabilitation feature set contains a plurality of rehabilitation features;

[0008] mapping the plurality of rehabilitation feature sets into the same vector space, and mining correlation information between rehabilitation features in different rehabilitation feature sets;

[0009] In the same rehabilitation process, two rehabilitation features are exhaustively extracted from each rehabilitation feature set as the first rehabilitation feature pair, forming multiple different first rehabilitation feature pairs;

[0010] In different rehabilitation processes, exhaustively extracting one rehabilitation feature from each of the two rehabilitation feature sets according to the association information as a second rehabilitation feature pair, to form a plurality of different second rehabilitation feature pairs;

[0011] A fusion feature set is constructed based on multiple pairs of the first and second rehabilitation features, and the fusion feature set is input into a preset rehabilitation status assessment model to output rehabilitation status assessment results of the orthopedic patient at different rehabilitation processes.

[0012] Preferably, the step of exhaustively extracting one rehabilitation feature from each of the two rehabilitation feature sets according to the association information as a second rehabilitation feature pair, to form a plurality of different second rehabilitation feature pairs, includes:

[0013] calculating the cosine distances between rehabilitation features in different rehabilitation feature sets based on the association information;

[0014] According to the cosine distance exhaustive method, a rehabilitation feature is extracted from each of the two rehabilitation feature sets as a second rehabilitation feature pair to form a plurality of different second rehabilitation feature pairs, wherein the cosine distance between the two rehabilitation features of the second rehabilitation feature pair is less than a preset cosine distance.

[0015] Preferably, the step of inputting the fusion feature set into a preset rehabilitation status assessment model and outputting the rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages includes:

[0016] The generator of the rehabilitation status evaluation model is used to analyze the input fusion feature set to generate the virtual rehabilitation status of the orthopedic patient at different rehabilitation processes;

[0017] Determine whether the virtual rehabilitation states of different rehabilitation processes are based on all key rehabilitation feature pairs in the fusion feature set;

[0018] If so, the discriminator of the rehabilitation state assessment model is used to compare the virtual rehabilitation states of different rehabilitation processes with the real rehabilitation states of the corresponding rehabilitation processes, and the virtual rehabilitation states whose similarity with the real rehabilitation states of the corresponding rehabilitation processes is greater than a preset similarity are screened out to obtain the target rehabilitation state of each rehabilitation process;

[0019] The target rehabilitation status of each rehabilitation process is evaluated respectively to generate rehabilitation status evaluation results of the orthopedic patient in different rehabilitation processes.

[0020] Preferably, the step of inputting the fusion feature set into a preset rehabilitation status assessment model and outputting the rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages includes:

[0021] Inputting the fused feature set into a preset rehabilitation status assessment model to generate a rehabilitation status assessment result for each first and second rehabilitation feature pair;

[0022] Check the consistency of rehabilitation status assessment results in the same rehabilitation process or between different rehabilitation processes, and evaluate the stability of rehabilitation status assessment results under different rehabilitation processes or different rehabilitation feature combinations, and output rehabilitation status assessment results that meet the consistency and stability requirements.

[0023] Preferably, mining correlation information between rehabilitation features in different rehabilitation feature sets includes:

[0024] The Pearson correlation coefficients between rehabilitation features in different rehabilitation feature sets were calculated respectively;

[0025] The corresponding association information is determined based on the Pearson correlation coefficients between the rehabilitation features in different rehabilitation feature sets.

[0026] Preferably, the acquisition of multimodal rehabilitation data of orthopedic patients after total knee replacement surgery at different rehabilitation stages includes:

[0027] Obtain training videos of knee flexion and extension exercises completed by orthopedic patients after total knee replacement at different stages of rehabilitation;

[0028] Divide the training video into multiple video frames, each video frame corresponds to a different part of the knee joint;

[0029] Calculate the mean and standard deviation of the gradient of each video frame respectively, set the corresponding adjustable parameters according to the location of the knee joint corresponding to each video frame, and add the mean value of the gradient of each video frame to the product of the corresponding standard deviation and the adjustable parameter to obtain the gradient value of each video frame;

[0030] Calculating the gradient difference between adjacent video frames in the training video according to the gradient values ​​of all video frames;

[0031] Performing edge detection on the training video to extract edge information of the knee joint area;

[0032] The gradient difference of adjacent video frames is fused with edge information, the amplitude of motion change is calculated based on the fused information, and video frames with motion change amplitude greater than a preset value are extracted as multimodal rehabilitation data.

[0033] Furthermore, after inputting the fusion feature set into a preset rehabilitation status assessment model and outputting the rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages, the method further includes:

[0034] determining an abnormal rehabilitation process with an abnormal rehabilitation state and abnormal characteristics of the abnormal rehabilitation process according to the rehabilitation state assessment result;

[0035] The rehabilitation program corresponding to the abnormal rehabilitation process is dynamically adjusted and optimized according to the abnormal characteristics to generate an optimized and adjusted target rehabilitation program.

[0036] The present application also provides a device for tracking the rehabilitation progress of orthopedic patients, comprising:

[0037] The acquisition module is used to obtain multimodal rehabilitation data of orthopedic patients at different rehabilitation stages after total knee replacement surgery;

[0038] a first extraction module, configured to perform feature extraction on the multimodal rehabilitation data to form a rehabilitation feature set for each rehabilitation process, wherein each rehabilitation feature set contains a plurality of rehabilitation features;

[0039] a mining module, configured to map the plurality of rehabilitation feature sets into the same vector space and mine association information between rehabilitation features in different rehabilitation feature sets;

[0040] a second extraction module, configured to exhaustively extract two rehabilitation features from each rehabilitation feature set as first rehabilitation feature pairs in the same rehabilitation process, thereby forming a plurality of different first rehabilitation feature pairs;

[0041] a third extraction module, configured to, in different rehabilitation processes, exhaustively extract one rehabilitation feature from each of two rehabilitation feature sets according to the association information as a second rehabilitation feature pair, to form a plurality of different second rehabilitation feature pairs;

[0042] A construction module is used to construct a fusion feature set based on multiple pairs of the first and second rehabilitation features, input the fusion feature set into a preset rehabilitation status assessment model, and output the rehabilitation status assessment results of the orthopedic patient in different rehabilitation processes.

[0043] The present application also provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for tracking the rehabilitation progress of orthopedic patients as described in any one of the above items is implemented.

[0044] The present application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any of the above-mentioned methods for tracking the rehabilitation progress of orthopedic patients are implemented.

[0045] The present application provides a method, device, storage medium and equipment for tracking the rehabilitation process of orthopedic patients. By extracting features from multimodal rehabilitation data, a rehabilitation feature set for each rehabilitation process is formed. In the same rehabilitation process, two rehabilitation features are exhaustively extracted from each rehabilitation feature set to form multiple different first rehabilitation feature pairs. In different rehabilitation processes, one rehabilitation feature is exhaustively extracted from each two rehabilitation feature sets based on the associated information to form multiple different second rehabilitation feature pairs. In order to integrate multimodal data, mine the relationship between different modal data, and mine the data relationship between the same or different rehabilitation processes, the patient's rehabilitation status in each rehabilitation process can be comprehensively and accurately evaluated. This is more scientific and reliable than relying solely on a certain type of data or simple observation and evaluation, allowing doctors to accurately grasp the patient's recovery status. In addition, timely and accurate understanding of the rehabilitation process can avoid over-rehabilitation or under-rehabilitation, allowing patients to more efficiently restore knee joint function, shorten the rehabilitation cycle, save medical resources, and improve the overall efficiency of orthopedic rehabilitation. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flowchart of a method for tracking the rehabilitation progress of orthopedic patients according to one embodiment of the present application;

[0047] Figure 2 This is a schematic block diagram of a device for tracking the rehabilitation progress of orthopedic patients according to one embodiment of the present application;

[0048] Figure 3 This is a schematic block diagram of the structure of an electronic device according to an embodiment of the present application.

[0049] The realization of the objectives, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0051] refer to Figure 1 As shown, in one embodiment, the present application provides a method for tracking the rehabilitation progress of an orthopedic patient, the method comprising:

[0052] S11. Obtain multimodal rehabilitation data of orthopedic patients at different stages of rehabilitation after total knee replacement;

[0053] S12. Extracting features from the multimodal rehabilitation data to form a rehabilitation feature set for each rehabilitation process, where each rehabilitation feature set contains multiple rehabilitation features.

[0054] S13, mapping the multiple rehabilitation feature sets into the same vector space, and mining correlation information between rehabilitation features in different rehabilitation feature sets;

[0055] S14. In the same rehabilitation process, exhaustively extract two rehabilitation features from each rehabilitation feature set as first rehabilitation feature pairs, forming multiple different first rehabilitation feature pairs;

[0056] S15. In different rehabilitation processes, exhaustively extracting one rehabilitation feature from each of the two rehabilitation feature sets according to the association information as a second rehabilitation feature pair, to form a plurality of different second rehabilitation feature pairs;

[0057] S16. Construct a fusion feature set based on the plurality of first and second rehabilitation feature pairs, input the fusion feature set into a preset rehabilitation status assessment model, and output rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages.

[0058] Among them, multimodal rehabilitation data may include the patient's physiological data (such as changes in vital signs such as heart rate and blood pressure), motor function data (such as knee joint range of motion, by measuring the angle of joint flexion and extension), imaging data (such as regularly taken knee joint X-rays or MRI images to observe bone and soft tissue recovery), and subjective perception data (such as the patient's own rating of pain intensity, subjective evaluation of knee joint function recovery, etc.). This multimodal data is collected in different stages of rehabilitation, such as the early (1-2 weeks), mid-term (3-6 weeks), and late (2-3 months) after surgery, in order to comprehensively track the patient's recovery.

[0059] Feature extraction is performed on multimodal rehabilitation data to form a rehabilitation feature set for each rehabilitation process. For example, for physiological data, feature extraction can be to extract the average heart rate and average blood pressure value in a stable state as physiological features; for motor function data, feature extraction can be the maximum motion angle of the knee joint in different directions, the speed change characteristics during movement, etc.; imaging data can extract morphological features such as bone healing (the range and density of callus growth), soft tissue swelling degree change characteristics, etc.; in subjective perception data, taking pain scores as an example, pain scores at different time points and score change trend characteristics can be extracted. The features extracted from various modal data for each rehabilitation process are summarized to form a corresponding rehabilitation feature set, and each set covers multiple features that reflect the rehabilitation status from different angles.

[0060] Through mathematical transformation or encoding, the rehabilitation feature sets of different rehabilitation processes are converted into a unified vector space that can be compared and analyzed. For example, principal component analysis can be used to represent the feature sets of different rehabilitation processes with corresponding coordinates in the same vector space. Then, the correlation information of rehabilitation features between different rehabilitation feature sets is mined, for example, the correlation between the improvement of knee joint range of motion in the early rehabilitation process and the recovery of muscle strength in the mid-term rehabilitation process, or the correlation between bone healing shown in early imaging and joint stability in the later stage, etc., so as to construct a comprehensive rehabilitation feature correlation network to help understand the key factors and mutual influence relationships in the entire rehabilitation process.

[0061] In the same rehabilitation process, two rehabilitation features are extracted from each rehabilitation feature set as the first rehabilitation feature pair, forming multiple different first rehabilitation feature pairs. For example, in the rehabilitation feature set of a specific rehabilitation process (assuming it is the mid-term postoperative stage), assuming there are 5 rehabilitation features, then these 5 rehabilitation features should be combined in pairs, exhausting all possible combinations of two features to form multiple groups of first rehabilitation feature pairs. For example, from rehabilitation features A, B, C, D, E, there will be multiple groups of first rehabilitation feature pairs such as (A, B), (A, C), (A, D)... (C, D), (C, E), (D, E), etc., to fully explore the correlation between the data of different rehabilitation features in the rehabilitation process, facilitating the subsequent analysis of the impact of feature interactions on the rehabilitation status within the same rehabilitation stage.

[0062] In different rehabilitation processes, based on the exhaustive association information, a rehabilitation feature is extracted from each of the two rehabilitation feature sets as the second rehabilitation feature pair, forming multiple different second rehabilitation feature pairs. For example, based on the association information between the rehabilitation features previously mined, such as a certain feature in the early stage and a certain feature in the middle stage, a feature is extracted from the corresponding rehabilitation feature sets in the early stage and the middle stage to form the second rehabilitation feature pair. Assuming that the early rehabilitation feature set has features X and Y, and the middle rehabilitation feature set has features M and N, if it is found in the early stage that X and M are associated, and Y and N are associated, then the second rehabilitation feature pairs such as (X, M) and (Y, N) are formed. Similarly, this cross-rehabilitation process feature combination based on association information is exhausted to analyze the connection between features in different rehabilitation stages, which facilitates the subsequent analysis of the impact of feature interactions on the rehabilitation status in different rehabilitation stages.

[0063] The first rehabilitation feature pairs obtained previously within the same rehabilitation process, as well as the second rehabilitation feature pairs between different rehabilitation processes, are combined to form a fused feature set. This fused feature set encompasses a large number of feature pairs extracted across different dimensions (both within the same process and between different processes). These feature pairs reflect various correlations in the patient's rehabilitation process from multiple perspectives, providing a more comprehensive representation of the patient's recovery status and richer information for subsequent evaluation.

[0064] Finally, the fused feature set is input into a pre-set rehabilitation status assessment model, which outputs the rehabilitation status assessment results for orthopedic patients at different stages of their rehabilitation. This rehabilitation status assessment model can be based on machine learning or deep learning, such as a neural network model. After training with a large amount of training data (feature set samples with known rehabilitation status), it can comprehensively analyze and output the patient's specific rehabilitation status at different stages of the rehabilitation process, such as whether the patient has recovered well, recovered moderately, or recovered poorly, based on the combination of various feature pairs in the input fused feature set after screening and comparison. This helps doctors better understand the patient's rehabilitation progress.

[0065] For example, suppose there is a patient who has undergone total knee replacement. Rehabilitation is divided into three main processes: early, mid-term, and late. During the rehabilitation process, multimodal data was collected. The data collected in the early stage include: physiological data (average heart rate 80 beats / min, average blood pressure 120 / 80 mmHg, etc.), knee joint range of motion measurement data (flexion and extension angle from 0° to 30°), and pain score (6 points, out of 10 points, the higher the score, the more severe the pain). The mid-term data include: physiological data (average heart rate 75 beats / min, blood pressure 115 / 75 mmHg, etc.), knee joint range of motion (can flex and extend to 60°), and pain score (3 points). Late data include: physiological data (average heart rate 70 beats / min, blood pressure 110 / 70 mmHg, etc.), knee joint range of motion (close to normal, flexion and extension can reach about 120°), and pain score (0 points).

[0066] Feature extraction is performed on these different modal data. For example, the maximum flexion and extension angle (30°) is extracted from the early knee range of motion data as a motor function feature, and the pain intensity (6 points) is extracted from the pain score as a subjective perception feature. This creates a rehabilitation feature set for each rehabilitation process. These feature sets are then mapped to the same vector space to mine for correlation information. For example, a correlation was found between early knee range of motion and mid-term pain reduction. Next, within the same rehabilitation process (e.g., the early stage), two features are exhaustively extracted from the rehabilitation feature set as a first rehabilitation feature pair, such as (early knee range of motion, early pain score). For different rehabilitation processes, a rehabilitation feature is extracted from each of the two rehabilitation feature sets based on the correlation information, such as (early knee range of motion, mid-term pain score). These first and second rehabilitation feature pairs are aggregated into a fused feature set and input into the rehabilitation status assessment model. Ultimately, the patient's rehabilitation status for each rehabilitation process is determined, such as early recovery, gradual improvement, and good recovery.

[0067] This embodiment extracts features from multimodal rehabilitation data to form a rehabilitation feature set for each rehabilitation process. In the same rehabilitation process, two rehabilitation features are exhaustively extracted from each rehabilitation feature set to form multiple different first rehabilitation feature pairs. In different rehabilitation processes, one rehabilitation feature is exhaustively extracted from each of the two rehabilitation feature sets based on the associated information to form multiple different second rehabilitation feature pairs. This integrates multimodal data, mines the relationship between different modal data, and mines the data relationship between the same or different rehabilitation processes. This can comprehensively and accurately assess the patient's rehabilitation status in each rehabilitation process, which is more scientific and reliable than relying solely on a single data or simple observation and assessment, allowing doctors to accurately grasp the patient's recovery status. In addition, timely and accurate understanding of the rehabilitation process can avoid over-rehabilitation or under-rehabilitation, allowing patients to more efficiently restore knee joint function, shorten the rehabilitation cycle, save medical resources, and improve the overall efficiency of orthopedic rehabilitation.

[0068] In one embodiment, the step of exhaustively extracting one rehabilitation feature from each of the two rehabilitation feature sets according to the association information as a second rehabilitation feature pair, thereby forming a plurality of different second rehabilitation feature pairs, including:

[0069] calculating the cosine distances between rehabilitation features in different rehabilitation feature sets based on the association information;

[0070] According to the cosine distance exhaustive method, a rehabilitation feature is extracted from each of the two rehabilitation feature sets as a second rehabilitation feature pair to form a plurality of different second rehabilitation feature pairs, wherein the cosine distance between the two rehabilitation features of the second rehabilitation feature pair is less than a preset cosine distance.

[0071] Cosine distance is a method for measuring the similarity between two vectors. It reflects the angle between them by calculating the ratio of the dot product of the two vectors to the product of their moduli. Values ​​closer to 1 indicate higher similarity (smaller angle), while values ​​closer to 0 indicate lower similarity (larger angle). This embodiment utilizes association information—that is, previously discovered relationships between rehabilitation features in different rehabilitation feature sets—to calculate the cosine distance between pairs of rehabilitation features in each rehabilitation feature set. This quantifies the associations between different rehabilitation features, thereby providing a basis for further screening of suitable second rehabilitation feature pairs.

[0072] Among them, the preset cosine distance is a set threshold value used to screen out rehabilitation feature pairs with sufficient differences in different rehabilitation feature sets. In every two rehabilitation feature sets (such as the early rehabilitation feature set and the mid-term rehabilitation feature set, etc.), based on the previously calculated cosine distance, all possible feature pairs consisting of one rehabilitation feature are extracted from each rehabilitation feature set. From the feature pairs, the feature pairs with a cosine distance less than the preset cosine distance are selected as the second rehabilitation feature pairs to ensure that the selected second rehabilitation feature pairs have more obvious differences in the feature sets of different rehabilitation processes, so as to avoid the features being too similar and unable to effectively reflect the changes in feature associations at different rehabilitation stages.

[0073] For example, let's continue with the rehabilitation scenario after total knee replacement surgery mentioned above. Assume that in the early rehabilitation feature set, there are rehabilitation feature A (a feature of knee joint range of motion) and rehabilitation feature B (a feature of pain score), and in the mid-term rehabilitation feature set, there are rehabilitation feature C (a feature of muscle strength) and rehabilitation feature D (a feature of the accuracy of rehabilitation training movements). The cosine distance of rehabilitation features between different rehabilitation feature sets is calculated based on the associated information. For example, the cosine distance between feature A in the early rehabilitation feature set and feature C in the mid-term rehabilitation feature set is 0.7, the cosine distance between feature A and feature D is 0.3, the cosine distance between feature B and feature C is 0.4, and the cosine distance between feature B and feature D is 0.6. The preset cosine distance is 0.5. Then, when screening, for the combination of early and mid-term rehabilitation feature sets, feature A and feature C (0.7>0.5) cannot form the second rehabilitation feature pair, and feature B and feature D (0.6>0.5) cannot form the second rehabilitation feature pair. However, feature A and feature D (0.3<0.5) and feature B and feature C (0.4<0.5) can form the second rehabilitation feature pair. In this way, the second rehabilitation feature pairs that meet the cosine distance conditions are screened from the feature sets of different rehabilitation processes and used for the subsequent construction of the fusion feature set. Ultimately, these second rehabilitation feature pairs that meet the conditions are combined with the first rehabilitation feature pairs in the same rehabilitation process to form a fusion feature set, which is input into the rehabilitation status assessment model to obtain accurate rehabilitation status assessment results.

[0074] By introducing a cosine distance calculation and screening mechanism, this embodiment can more specifically screen for pairs of rehabilitation features with sufficient differentiation across different rehabilitation processes, avoiding excessive redundancy or similarity between features. This allows the fused feature set to better highlight significant changes and differences between features at different rehabilitation stages, thereby improving the accuracy and reliability of the rehabilitation status assessment model. Furthermore, feature pairs with cosine distances less than the preset cosine distance reflect significant correlations in features across different rehabilitation processes, making the assessment results more reasonable and reliable, and more clearly demonstrating the mutual influence and changing relationships between different features across the rehabilitation process, helping doctors and other professionals better understand and interpret the rehabilitation status assessment conclusions.

[0075] Preferably, the step of inputting the fusion feature set into a preset rehabilitation status assessment model and outputting the rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages specifically includes:

[0076] The generator of the rehabilitation status evaluation model is used to analyze the input fusion feature set to generate the virtual rehabilitation status of the orthopedic patient at different rehabilitation processes;

[0077] Determine whether the virtual rehabilitation states of different rehabilitation processes are based on all key rehabilitation feature pairs in the fusion feature set;

[0078] If so, the discriminator of the rehabilitation state assessment model is used to compare the virtual rehabilitation states of different rehabilitation processes with the real rehabilitation states of the corresponding rehabilitation processes, and the virtual rehabilitation states whose similarity with the real rehabilitation states of the corresponding rehabilitation processes is greater than a preset similarity are screened out to obtain the target rehabilitation state of each rehabilitation process;

[0079] The target rehabilitation status of each rehabilitation process is evaluated respectively to generate rehabilitation status evaluation results of the orthopedic patient in different rehabilitation processes.

[0080] The generator is a key component of the rehabilitation status assessment model, possessing powerful data processing and pattern generation capabilities. Once the fused feature set is input into the generator, it conducts an in-depth analysis of the various feature pairs within the fused feature set, including feature correlations within the same rehabilitation process and between different rehabilitation processes. Using its internal deep learning-based generative adversarial network, the generator constructs possible rehabilitation states for orthopedic patients at different stages of their rehabilitation. These are known as virtual rehabilitation states. These virtual rehabilitation states encompass various scenarios of potential knee function recovery, from the early to late stages of rehabilitation. For example, in the early stages, a virtual state might manifest as a slight increase in knee range of motion and some relief of pain. In the mid-stage, this might be a further increase in range of motion and the beginnings of muscle strength recovery. The generator generates as many possible rehabilitation scenarios as possible.

[0081] This embodiment can identify and mark all key rehabilitation feature pairs in the fusion feature set, and determine whether the virtual rehabilitation states of different rehabilitation processes are based on all key rehabilitation feature pairs in the fusion feature set. For example, each key rehabilitation feature pair in the fusion feature set is perturbed (such as increasing or decreasing its value within a certain range), and then the virtual rehabilitation state is regenerated and the changes in the results are observed. If perturbing a key feature pair causes a significant change in the virtual rehabilitation state, it means that the feature pair plays an important role in the model and the virtual rehabilitation state is indeed based on the feature pair; conversely, if the virtual rehabilitation state hardly changes after the perturbation, it may mean that the key feature pair is not fully relied upon. Therefore, by performing perturbation tests on all key feature pairs in turn, it can be determined which key feature pairs are effectively utilized in generating the virtual rehabilitation state.

[0082] If not, the virtual recovery state is regenerated, supplemented, improved or deleted.

[0083] If so, the discriminator of the rehabilitation state assessment model compares the virtual rehabilitation states of different rehabilitation processes with the actual rehabilitation states of the corresponding rehabilitation processes. It selects virtual rehabilitation states whose similarity to the actual rehabilitation states of the corresponding rehabilitation processes exceeds a preset similarity threshold, thereby obtaining the target rehabilitation state for each rehabilitation process. The discriminator is another key component of the rehabilitation state assessment model, and its primary function is to perform comparison and screening. Each rehabilitation process has a corresponding actual rehabilitation state. These actual states can be determined through professional evaluations by doctors, clinical examination results, and other means and stored in a database. The discriminator compares the virtual rehabilitation states generated by the generator with the actual rehabilitation states of the corresponding rehabilitation processes one by one and calculates the similarity between them. This similarity can be calculated based on various factors, such as the numerical similarity of knee range of motion, pain intensity, muscle strength, and the consistency of their change trends. Only virtual rehabilitation states with a similarity greater than a preset similarity threshold are selected as target rehabilitation states. The preset similarity threshold is a threshold set based on practical needs and experience to ensure that the selected target rehabilitation state is sufficiently close to the actual state to ensure the reliability of subsequent assessment results.

[0084] The rehabilitation status assessment model will further evaluate the target rehabilitation status selected for each rehabilitation process. The assessment is mainly based on the various characteristic indicators contained in the target rehabilitation status, such as whether the knee joint range of motion has reached the expected range of the rehabilitation process, whether the degree of pain relief meets the standard, and the recovery of muscle strength. These indicators are combined to determine the patient's specific rehabilitation status assessment results for the rehabilitation process, such as "good rehabilitation progress", "general rehabilitation progress but in line with expectations", or "slow rehabilitation progress requires strengthened intervention". Ultimately, detailed rehabilitation status assessment results for orthopedic patients in different rehabilitation processes are obtained, providing accurate reference for doctors to adjust or formulate rehabilitation plans.

[0085] This embodiment, through the collaborative work of the generator and discriminator, first generates multiple possible virtual rehabilitation states, then rigorously compares and screens them against the real states. This effectively avoids assessment bias caused by small amounts of data or a single assessment method, making the final rehabilitation state assessment results more accurate and reliable. Simultaneously, by separately assessing the target rehabilitation state for each rehabilitation process, it can delve deeper into the patient's recovery details at each specific stage, providing a more refined understanding of rehabilitation progress. This helps to promptly identify problems and adjust rehabilitation plans, improving the quality and effectiveness of rehabilitation treatment and promoting better recovery of knee joint function.

[0086] In one embodiment, the inputting of the fusion feature set into a preset rehabilitation status assessment model to output the rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages includes:

[0087] Inputting the fused feature set into a preset rehabilitation status assessment model to generate a rehabilitation status assessment result for each first and second rehabilitation feature pair;

[0088] Check the consistency of rehabilitation status assessment results in the same rehabilitation process or between different rehabilitation processes, and evaluate the stability of rehabilitation status assessment results under different rehabilitation processes or different rehabilitation feature combinations, and output rehabilitation status assessment results that meet the consistency and stability requirements.

[0089] In this embodiment, the fused feature set is input into a pre-set rehabilitation status assessment model. Based on the feature pairs in the fused feature set, the model generates rehabilitation status assessment results for each feature pair. For example, the model might assess a first rehabilitation feature pair as "good" and a second rehabilitation feature pair as "needing attention."

[0090] Collect the evaluation results for all pairs of first and second rehabilitation features. For different pairs of first rehabilitation features within the same rehabilitation process, check whether their evaluation results are consistent. For example, if the results for most first rehabilitation feature pairs are "good," but some are "fair," further analysis of these inconsistent results is necessary.

[0091] For pairs of second rehabilitation features across different rehabilitation processes, check whether their assessment results are logically consistent. For example, if the result in the early rehabilitation process is "good" but suddenly changes to "poor" in the middle rehabilitation process, analyze whether this change is reasonable.

[0092] Secondly, the stability of the rehabilitation status assessment results at different time points was evaluated. If the assessment results of a feature pair remain stable at multiple consecutive time points, the result is considered to have high stability.

[0093] For different feature combinations within the same rehabilitation process or across different rehabilitation processes, analyze whether their assessment results are stable. If the assessment results of different feature combinations are consistent and stable, the credibility of the assessment results is increased.

[0094] Based on the consistency and stability assessment, valid and reliable rehabilitation status assessment results are screened out, and the screened assessment results are integrated into the final rehabilitation status assessment report, which reflects the patient's rehabilitation status in different rehabilitation processes.

[0095] This application ensures the accuracy and reliability of rehabilitation status assessment results by checking consistency and stability, and ensures the logical consistency of assessment results between different rehabilitation processes, thereby promoting the continuity and coherence of the rehabilitation process.

[0096] In one embodiment, mining correlation information between rehabilitation features in different rehabilitation feature sets includes:

[0097] The Pearson correlation coefficients between rehabilitation features in different rehabilitation feature sets were calculated respectively;

[0098] The corresponding association information is determined based on the Pearson correlation coefficients between the rehabilitation features in different rehabilitation feature sets.

[0099] Rehabilitation feature pairs were selected from different rehabilitation feature sets for analysis. For example, knee range of motion (feature A) and pain score (feature B) were selected from the early rehabilitation feature set, and muscle strength (feature C) and rehabilitation training accuracy (feature D) were selected from the mid-term rehabilitation feature set.

[0100] The mean and standard deviation of each rehabilitation feature were calculated, and the covariance between two rehabilitation features was calculated. The Pearson correlation coefficient was calculated using the covariance and standard deviation. The formula is as follows:

[0101]

[0102] Among them, σ x and σ yare the standard deviations of rehabilitation features x and y, respectively. Cov(x, y) is the covariance between the two rehabilitation features. r is the Pearson correlation coefficient, which ranges from -1 to 1. The closer the absolute value is to 1, the stronger the correlation. A positive value indicates a positive correlation, and a negative value indicates a negative correlation.

[0103] All calculated Pearson correlation coefficients are organized into a matrix. The rows and columns of the matrix represent different rehabilitation features, and the elements in the matrix represent the Pearson correlation coefficient between two rehabilitation features. The reference is as follows:

[0104]

[0105] The correlation coefficient matrix was used to analyze the associations between different rehabilitation features. For example, if the Pearson correlation coefficient between knee range of motion (Feature A) in the early rehabilitation feature set and muscle strength (Feature C) in the mid-term rehabilitation feature set was 0.75, it indicated a strong positive correlation between the two. If the Pearson correlation coefficient between pain score (Feature B) in the early rehabilitation feature set and rehabilitation training accuracy (Feature D) in the mid-term rehabilitation feature set was -0.60, it indicated a strong negative correlation between the two.

[0106] In addition, visualization tools such as heat maps can be used to display the correlation coefficient matrix, intuitively presenting the association information between different rehabilitation features, or to use color depth to indicate the size and direction of the correlation coefficient, making it easier to quickly identify strongly correlated and weakly correlated feature pairs.

[0107] This embodiment calculates the Pearson correlation coefficient to explore the correlation information between different rehabilitation features, improve the representativeness of the features, and more comprehensively reflect the patient's recovery status. The correlation coefficient matrix and heat map can intuitively present the correlation information between different rehabilitation features, enhancing the interpretability of the features.

[0108] In one embodiment, obtaining multimodal rehabilitation data of orthopedic patients undergoing total knee replacement at different rehabilitation stages includes:

[0109] Obtain training videos of knee flexion and extension exercises completed by orthopedic patients after total knee replacement at different stages of rehabilitation;

[0110] Divide the training video into multiple video frames, each video frame corresponds to a different part of the knee joint;

[0111] Calculate the mean and standard deviation of the gradient of each video frame respectively, set the corresponding adjustable parameters according to the location of the knee joint corresponding to each video frame, and add the mean value of the gradient of each video frame to the product of the corresponding standard deviation and the adjustable parameter to obtain the gradient value of each video frame;

[0112] Calculating the gradient difference between adjacent video frames in the training video according to the gradient values ​​of all video frames;

[0113] Performing edge detection on the training video to extract edge information of the knee joint area;

[0114] The gradient difference of adjacent video frames is fused with edge information, the amplitude of motion change is calculated based on the fused information, and video frames with motion change amplitude greater than a preset value are extracted as multimodal rehabilitation data.

[0115] This embodiment collects videos of knee flexion and extension training completed by patients at different rehabilitation stages, and divides the training videos into multiple video frames, each of which corresponds to a different part of the knee joint, so that each component of the knee joint can be analyzed one by one.

[0116] For each video frame, the mean and standard deviation of its gradient are calculated to understand the degree of image change in each video frame. The mean of the gradient reflects the overall change trend, while the standard deviation shows the discrete degree of the change.

[0117] The gradient value of each video frame is dynamically determined based on the calculated mean and standard deviation of the gradient. This dynamic adjustment helps to more accurately capture the motion characteristics of the knee joint. For example, the adjustable parameters corresponding to each video frame can be set according to the location of the knee joint in each video frame. The standard deviation of the gradient of each video frame is calculated and multiplied by the corresponding adjustable parameter. Finally, the mean value of the gradient of each video frame is added to the corresponding product to obtain the gradient value of each video frame. The specific formula includes the following:

[0118] L f =μ f +k*σ f ;

[0119] Among them, the L f is the gradient value of the video frame f, the μ f is the average value of the gradient of the video frame f, the σ f is the standard deviation of the gradient of the video frame f, and k is an adjustable parameter of the video frame f. Usually, a suitable value is selected according to the position of the corresponding knee joint, for example, k=1 or k=2.

[0120] This embodiment can set adjustable parameters according to the location of the knee joint corresponding to each video frame, which can make the processing process better adapt to the specific characteristics and changes of the knee joint in different video frames. Because the gradient characteristics of the knee joint will be different in different motion states, angles and positions, this targeted parameter setting can improve the accuracy of capturing and analyzing knee joint characteristics. In addition, by calculating the standard deviation of the gradient of each video frame and the product of the corresponding adjustable parameter, and adding it to the average value of the gradient to obtain the final gradient value, it is equivalent to adjusting the dynamic range of the gradient of each video frame, which helps to highlight the important characteristics and change information of the knee joint in the video frame, so that the video frames with key details about the knee joint can be accurately extracted later. For example, in the motion analysis of the knee joint, the friction of the joint surface and the stretching of the ligaments can be more clearly captured. Since certain subtle changes in the knee joint in the video sequence are crucial for accurate rehabilitation assessment, motion analysis, etc., this embodiment can make these details more fully reflected in the gradient value, improving the accuracy and reliability of the subsequent extraction of video frames reflecting the state of the knee joint, so as to optimize the acquisition of rehabilitation data.

[0121] Next, the gradient difference between adjacent video frames is calculated to capture the changes between video frames, that is, the dynamic changes of knee flexion and extension movements.

[0122] Edge detection is performed on the training video to extract edge information of the knee joint area. Edge detection helps highlight the contours of the knee joint and enhances image analysis. The gradient difference between adjacent video frames is fused with the edge information. The fused information is used to calculate the amplitude of motion changes. This combination of gradient changes and edge information allows for a more comprehensive assessment of knee joint motion.

[0123] Finally, video frames with movement changes greater than a preset value are extracted as multimodal rehabilitation data. This key frame represents the most important moment of change in the knee flexion and extension movement, thereby obtaining a set of detailed, multi-dimensional rehabilitation data, which provides a solid foundation for subsequent rehabilitation status assessment and improves the efficiency and accuracy of rehabilitation data acquisition.

[0124] In one embodiment, after inputting the fused feature set into a preset rehabilitation status assessment model to output the rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages, the method further includes:

[0125] determining an abnormal rehabilitation process with an abnormal rehabilitation state and abnormal characteristics of the abnormal rehabilitation process according to the rehabilitation state assessment result;

[0126] The rehabilitation program corresponding to the abnormal rehabilitation process is dynamically adjusted and optimized according to the abnormal characteristics to generate an optimized and adjusted target rehabilitation program.

[0127] For the evaluation results of each rehabilitation process, a corresponding normal range or standard model can be set. The normal range or standard model can be obtained based on statistical analysis of a large amount of clinical data, or can be formulated by experts based on experience.

[0128] Compare the actual assessment results with the normal range or standard pattern. For example, in terms of knee range of motion, if the assessment results of a rehabilitation process show that the knee range of motion is significantly lower than the lower limit of the normal range for that process, it can be determined that the process is abnormal.

[0129] Record and identify specific abnormal characteristics of the abnormal rehabilitation process, that is, key indicators or factors that lead to abnormal rehabilitation status, such as insufficient knee joint range of motion, slow muscle strength recovery, and insignificant pain relief.

[0130] Establish a mapping relationship between abnormal characteristics and rehabilitation program adjustment strategies. For example, if the abnormal characteristic is insufficient knee joint range of motion, the corresponding adjustment strategy may be to increase the intensity or frequency of knee flexion and extension exercises.

[0131] Based on the specific characteristics of the abnormality, a matching adjustment strategy is selected from a preset rehabilitation program library. The rehabilitation program library can include a variety of different combinations of rehabilitation training methods, intensities, frequencies, etc.

[0132] Optimization algorithms are used to dynamically adjust rehabilitation plans, generating a more targeted and effective target rehabilitation plan to help patients recover better. For example, optimization algorithms such as genetic algorithms and simulated annealing algorithms are used, with improvement of abnormal characteristics as the objective function, to optimize the combination of various parameters in the rehabilitation plan.

[0133] This embodiment can better meet the patient's individualized rehabilitation needs, improve rehabilitation effects, promote patients to recover knee joint function faster, and promptly detect and deal with abnormal situations in the rehabilitation process, avoiding the risk of rehabilitation stagnation or worsening of the condition due to unsuitable rehabilitation plans, thereby ensuring the safety of rehabilitation training.

[0134] Preferably, this embodiment utilizes blockchain technology to encrypt and store multimodal rehabilitation data and rehabilitation status assessment results, preventing data tampering during storage and transmission. Specifically, the AES algorithm is used to encrypt multimodal rehabilitation data and rehabilitation status assessment results, ensuring data security during storage and transmission. AES is a symmetric encryption algorithm with high encryption strength and speed, making it suitable for encrypting large amounts of data.

[0135] Then, a pair of public and private keys is generated for each patient. The private key is kept by the patient himself, and the public key can be made public. When encrypting data, the patient's public key is used to encrypt the data, so that only the patient with the corresponding private key can decrypt and view their own data.

[0136] In this embodiment, consortium chains or private chains can be used as the blockchain type for storing multimodal rehabilitation data and rehabilitation status assessment results. Consortium chains are suitable for sharing data between multiple medical institutions or rehabilitation centers, while private chains are suitable for data storage and management within a single medical institution.

[0137] At the same time, the blockchain data structure is designed, including a block header and a block body. The block header contains information such as the hash value of the previous block, the hash value of the current block, and a timestamp, while the block body stores the encrypted multimodal rehabilitation data, the hash value of the rehabilitation status assessment results, and related metadata.

[0138] A hash value is calculated for the encrypted multimodal rehabilitation data and rehabilitation status assessment results. This unique and tamper-proof hash value can be used to verify the integrity and authenticity of the data. The calculated hash value and associated metadata (such as data type, data source, and timestamp) are stored in a blockchain block. Simultaneously, the encrypted data is stored in an off-chain distributed storage system to address the limited storage capacity of the blockchain. A mapping between the on-chain hash value and the off-chain stored data is established on the blockchain, enabling rapid location and retrieval of the corresponding off-chain data when needed.

[0139] When a user requests access to multimodal rehabilitation data or rehabilitation status assessment results, the system authenticates and verifies their identity. Only authenticated and authorized users can decrypt and view the data. After accessing the off-chain data, users can verify it using the hash value on the blockchain to ensure it has not been tampered with during storage and transmission.

[0140] Reference Figure 2 As shown, an embodiment of the present application further provides a device for tracking the rehabilitation progress of orthopedic patients, the device comprising:

[0141] An acquisition module 11 is used to acquire multimodal rehabilitation data of orthopedic patients at different rehabilitation stages after total knee replacement surgery;

[0142] a first extraction module 12, configured to perform feature extraction on the multimodal rehabilitation data to form a rehabilitation feature set for each rehabilitation process, wherein each rehabilitation feature set contains a plurality of rehabilitation features;

[0143] A mining module 13 is configured to map the plurality of rehabilitation feature sets into the same vector space and mine association information between rehabilitation features in different rehabilitation feature sets;

[0144] The second extraction module 14 is configured to exhaustively extract two rehabilitation features from each rehabilitation feature set as first rehabilitation feature pairs in the same rehabilitation process, thereby forming a plurality of different first rehabilitation feature pairs;

[0145] a third extraction module 15, configured to exhaustively extract, according to the association information, one rehabilitation feature from each of the two rehabilitation feature sets as a second rehabilitation feature pair in different rehabilitation processes, to form a plurality of different second rehabilitation feature pairs;

[0146] The construction module 16 is used to construct a fusion feature set based on multiple pairs of the first and second rehabilitation features, input the fusion feature set into a preset rehabilitation status assessment model, and output the rehabilitation status assessment results of the orthopedic patient in different rehabilitation processes.

[0147] As described above, it can be understood that the various components of the orthopedic patient rehabilitation progress tracking device proposed in this application can realize the functions of any of the orthopedic patient rehabilitation progress tracking methods described above, and the specific structure will not be repeated.

[0148] Reference Figure 3 As shown, an electronic device is also provided in the embodiment of the present application, and its internal structure can be as follows Figure 3 As shown. The electronic device includes a processor, a memory, a network interface and a database connected via a system bus. The processor designed for the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a storage medium and an internal memory. The storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the storage medium. The database of the electronic device is used to store relevant data of the method for tracking the rehabilitation progress of orthopedic patients. The network interface of the electronic device is used to communicate with an external electronic device via a network connection. When the computer program is executed by the processor, a method for tracking the rehabilitation progress of orthopedic patients is implemented.

[0149] In one embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a method for tracking the rehabilitation progress of orthopedic patients is implemented.

[0150] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0151] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0152] The above description is only a preferred embodiment of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for tracking the rehabilitation progress of orthopedic patients, characterized in that: include: To obtain multimodal rehabilitation data of orthopedic patients at different stages of rehabilitation after total knee replacement; performing feature extraction on the multimodal rehabilitation data to form a rehabilitation feature set for each rehabilitation process, wherein each rehabilitation feature set contains a plurality of rehabilitation features; mapping the plurality of rehabilitation feature sets into the same vector space, and mining correlation information between rehabilitation features in different rehabilitation feature sets; In the same rehabilitation process, two rehabilitation features are exhaustively extracted from each rehabilitation feature set as the first rehabilitation feature pair, forming multiple different first rehabilitation feature pairs; In different rehabilitation processes, exhaustively extracting one rehabilitation feature from each of the two rehabilitation feature sets according to the association information as a second rehabilitation feature pair, to form a plurality of different second rehabilitation feature pairs; A fusion feature set is constructed based on multiple pairs of the first and second rehabilitation features, and the fusion feature set is input into a preset rehabilitation status assessment model to output rehabilitation status assessment results of the orthopedic patient at different rehabilitation processes.

2. The method according to claim 1, characterized in that The step of exhaustively extracting one rehabilitation feature from each of the two rehabilitation feature sets according to the association information as a second rehabilitation feature pair to form a plurality of different second rehabilitation feature pairs includes: calculating the cosine distances between rehabilitation features in different rehabilitation feature sets based on the association information; According to the cosine distance exhaustive method, a rehabilitation feature is extracted from each of the two rehabilitation feature sets as a second rehabilitation feature pair to form a plurality of different second rehabilitation feature pairs, wherein the cosine distance between the two rehabilitation features of the second rehabilitation feature pair is less than a preset cosine distance.

3. The method according to claim 1, characterized in that The step of inputting the fusion feature set into a preset rehabilitation status assessment model and outputting the rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages includes: The generator of the rehabilitation status evaluation model is used to analyze the input fusion feature set to generate the virtual rehabilitation status of the orthopedic patient at different rehabilitation processes; Determine whether the virtual rehabilitation states of different rehabilitation processes are based on all key rehabilitation feature pairs in the fusion feature set; If so, the discriminator of the rehabilitation state assessment model is used to compare the virtual rehabilitation states of different rehabilitation processes with the real rehabilitation states of the corresponding rehabilitation processes, and the virtual rehabilitation states whose similarity with the real rehabilitation states of the corresponding rehabilitation processes is greater than a preset similarity are screened out to obtain the target rehabilitation state of each rehabilitation process; The target rehabilitation status of each rehabilitation process is evaluated respectively to generate rehabilitation status evaluation results of the orthopedic patient in different rehabilitation processes.

4. The method according to claim 1, wherein The step of inputting the fusion feature set into a preset rehabilitation status assessment model and outputting the rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages includes: Inputting the fused feature set into a preset rehabilitation status assessment model to generate a rehabilitation status assessment result for each first and second rehabilitation feature pair; Check the consistency of rehabilitation status assessment results in the same rehabilitation process or between different rehabilitation processes, and evaluate the stability of rehabilitation status assessment results under different rehabilitation processes or different rehabilitation feature combinations, and output rehabilitation status assessment results that meet the consistency and stability requirements.

5. The method according to claim 1, wherein The mining of association information between rehabilitation features in different rehabilitation feature sets includes: The Pearson correlation coefficients between rehabilitation features in different rehabilitation feature sets were calculated respectively; The corresponding association information is determined based on the Pearson correlation coefficients between the rehabilitation features in different rehabilitation feature sets.

6. The method according to claim 1, characterized in that The multimodal rehabilitation data of orthopedic patients after total knee replacement at different rehabilitation stages are obtained, including: Obtain training videos of knee flexion and extension exercises completed by orthopedic patients after total knee replacement at different stages of rehabilitation; Divide the training video into multiple video frames, each video frame corresponds to a different part of the knee joint; Calculate the mean and standard deviation of the gradient of each video frame respectively, set the corresponding adjustable parameters according to the location of the knee joint corresponding to each video frame, and add the mean value of the gradient of each video frame to the product of the corresponding standard deviation and the adjustable parameter to obtain the gradient value of each video frame; Calculating the gradient difference between adjacent video frames in the training video according to the gradient values ​​of all video frames; Performing edge detection on the training video to extract edge information of the knee joint area; The gradient difference of adjacent video frames is fused with edge information, the amplitude of motion change is calculated based on the fused information, and video frames with motion change amplitude greater than a preset value are extracted as multimodal rehabilitation data.

7. The method according to claim 1, characterized in that After inputting the fusion feature set into a preset rehabilitation status assessment model and outputting the rehabilitation status assessment results of the orthopedic patient at different rehabilitation stages, the method further includes: determining an abnormal rehabilitation process with an abnormal rehabilitation state and abnormal characteristics of the abnormal rehabilitation process according to the rehabilitation state assessment result; The rehabilitation program corresponding to the abnormal rehabilitation process is dynamically adjusted and optimized according to the abnormal characteristics to generate an optimized and adjusted target rehabilitation program.

8. A device for tracking the rehabilitation progress of orthopedic patients, characterized in that: include: The acquisition module is used to obtain multimodal rehabilitation data of orthopedic patients at different rehabilitation stages after total knee replacement surgery; a first extraction module, configured to perform feature extraction on the multimodal rehabilitation data to form a rehabilitation feature set for each rehabilitation process, wherein each rehabilitation feature set contains a plurality of rehabilitation features; a mining module, configured to map the plurality of rehabilitation feature sets into the same vector space and mine association information between rehabilitation features in different rehabilitation feature sets; a second extraction module, configured to exhaustively extract two rehabilitation features from each rehabilitation feature set as first rehabilitation feature pairs in the same rehabilitation process, thereby forming a plurality of different first rehabilitation feature pairs; a third extraction module, configured to, in different rehabilitation processes, exhaustively extract one rehabilitation feature from each of two rehabilitation feature sets according to the association information as a second rehabilitation feature pair, to form a plurality of different second rehabilitation feature pairs; A construction module is used to construct a fusion feature set based on multiple pairs of the first and second rehabilitation features, input the fusion feature set into a preset rehabilitation status assessment model, and output the rehabilitation status assessment results of the orthopedic patient in different rehabilitation processes.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for tracking the rehabilitation progress of an orthopedic patient according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: processor; Memory; The memory stores a computer program, and the processor implements the method for tracking the rehabilitation progress of orthopedic patients according to any one of claims 1 to 7 when executing the computer program.

Images