Training program analysis system for renal rehabilitation
Through the user sample data analysis and distillation learning technology of the kidney rehabilitation training program analysis system, the problem of relying on doctor experience and lack of systematic analysis in the existing technology is solved, and the accurate evaluation and optimization of the kidney rehabilitation program is achieved, and the application of rehabilitation effect is improved. Sex and scientific.
Patent Information
- Application Number
- CN202411181813.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-08-27
AI Technical Summary
The existing renal rehabilitation training program relies on doctor experience and lacks systematic data analysis, resulting in inaccurate rehabilitation effects, poor applicability of the program, and overall optimization training may cause cross-interference, making it difficult to adjust for individual differences.
Provides a training program analysis system for renal rehabilitation, including training program library access, sample collection, utility analysis, program division and distillation learning modules. Through user sample data analysis and distillation learning technology, accurately evaluate and optimize rehabilitation programs to avoid crossovers Interference to ensure the targeted and scientific nature of the program.
Accurate evaluation and optimization of renal rehabilitation plans have been achieved, the applicability and scientificity of the rehabilitation plans have been improved, ensuring that patients get the most suitable rehabilitation plans, and significantly improving the reliability and efficiency of rehabilitation results.
Smart Images

Figure CN118711751B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field related to medical care information processing, and in particular to a training program analysis system for kidney rehabilitation. Background Art
[0002] Existing renal rehabilitation training programs are usually developed based on the doctor's experience and personal situation. However, in actual applications, due to individual differences among patients, complexity of the disease and other factors, there are significant differences in the rehabilitation effects of different programs. There is a lack of systematic data analysis support, and it is difficult to quantify the rehabilitation effects. At the same time, the adaptability and effectiveness vary greatly among different patients. The renal rehabilitation training program lacks specificity, and it is difficult to adjust the rehabilitation program in time to adapt to the patient's changes.
[0003] In summary, the existing technologies have technical problems such as reliance on expert experience, inaccurate evaluation of the effectiveness of rehabilitation programs, kidney rehabilitation programs are mostly overall optimization training, cross-interference caused by mixed training with high-performing programs, and error-prone training program analysis process. Summary of the invention
[0004] This application provides a training program analysis system for renal rehabilitation, aiming to solve the technical problems in the prior art, such as reliance on expert experience, inaccurate evaluation of the effectiveness of rehabilitation programs, the fact that most renal rehabilitation programs are overall optimization training, cross-interference caused by mixed training with high-performance programs, and error-prone training program analysis process.
[0005] In view of the above problems, the technical solution to implement this application is:
[0006] The present application provides a training program analysis system for renal rehabilitation, which includes: a training program library access module for accessing a rehabilitation training program library, wherein the rehabilitation training program library stores a plurality of renal rehabilitation programs; a training sample collection module for collecting user group training samples for any rehabilitation program in the rehabilitation training program library to obtain a user sample data set, wherein the user sample data set includes a user rehabilitation data change sample, a user rehabilitation recovery effect sample, and a user training feedback sample; a utility analysis module for performing utility analysis on each renal rehabilitation program according to the user sample data set, and outputting a plurality of program failure indicators corresponding to the plurality of renal rehabilitation programs; a rehabilitation program division module; A block is used to divide the rehabilitation training program library according to the multiple program failure indicators to obtain a first type of renal rehabilitation program and a second type of renal rehabilitation program, wherein the first type of renal rehabilitation program is a set of programs whose program failure indicators are less than or equal to a preset failure threshold, and the second type of renal rehabilitation program is a set of programs whose program failure indicators are greater than the preset failure threshold; a distillation learning training module is used to take the preset failure threshold as the training target, perform distillation learning training on the second type of renal rehabilitation program based on the first type of renal rehabilitation program, and update the second type of renal rehabilitation program; a program delivery module is used to deliver the first type of renal rehabilitation program and the updated second type of renal rehabilitation program.
[0007] In summary, one or more technical solutions provided in this application solve the technical problems of reliance on expert experience, inaccurate evaluation of the effectiveness of rehabilitation programs, cross-interference caused by mixed training of renal rehabilitation programs with excellent performing programs, and error-prone training program analysis process. It realizes the collection of user training samples and utility analysis, accurately divides programs with different performances, and independently optimizes only renal rehabilitation programs with poor performance to avoid cross-interference, ensures the focus and accuracy of the optimization process, improves optimization efficiency, accurately evaluates the effectiveness of each renal rehabilitation program, ensures that the renal rehabilitation program library is always in the best state, provides patients with the most suitable rehabilitation program, ensures the applicability of renal rehabilitation programs, and significantly improves the scientificity and effectiveness of renal rehabilitation programs. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] Figure 1 A schematic diagram of a training program analysis system for renal rehabilitation is provided for this application;
[0009] Figure 2 A flowchart of the execution steps of updating the second type of renal rehabilitation program of the distillation learning training module in the training program analysis system for renal rehabilitation is provided for this application.
[0010] Explanation of the accompanying drawings: training program library access module M100, training sample collection module M200, utility analysis module M300, rehabilitation program division module M400, distillation learning training module M500, program delivery module M600. DETAILED DESCRIPTION
[0011] Example
[0012] The present application is described in detail below with reference to the accompanying drawings. Figure 1 As shown, the present application provides a training program analysis system for renal rehabilitation, which includes:
[0013] The training program library access module M100 is used to access the rehabilitation training program library, wherein the rehabilitation training program library stores multiple kidney rehabilitation programs; the training sample collection module M200 is used to collect user group training samples for any rehabilitation program in the rehabilitation training program library to obtain a user sample data set, wherein the user sample data set includes user rehabilitation data change samples, user rehabilitation recovery effect samples and user training feedback samples.
[0014] The embodiment of the present application proposes a universal rehabilitation training program, including access to a rehabilitation training program library, wherein the rehabilitation training program library stores multiple renal rehabilitation programs, and the multiple renal rehabilitation programs are the basis for subsequent analysis, optimization and delivery. For any renal rehabilitation program in the rehabilitation training program library, it is necessary to collect user group training samples; the purpose of collecting user group training samples is to obtain a comprehensive user sample data set, and the user sample data set is used for subsequent utility analysis of the rehabilitation program. The user sample data set includes user rehabilitation data change samples, user rehabilitation recovery effect samples and user training feedback samples. Specifically, the user rehabilitation data change samples record the various items of the patient during the process of accepting the rehabilitation program. Changes in physiological indicators (such as renal function indicators, blood pressure, blood sugar, etc.) can intuitively reflect the degree of improvement of the patient's physical condition by the rehabilitation program; user rehabilitation recovery effect samples are used to evaluate the patient's overall rehabilitation recovery effect, including the patient's symptom improvement, the degree of improvement in quality of life, and whether the expected rehabilitation goals are achieved, etc., which are generally obtained through questionnaires, doctor evaluations or patient self-evaluations; user training feedback samples, that is, patients' training feedback, are important indicators for evaluating the acceptance and satisfaction of rehabilitation programs, including patients' acceptance of rehabilitation programs, their feelings during training, difficulties encountered, and suggestions, etc. By collecting and analyzing feedback, we can understand the patients' real needs and expectations for the rehabilitation program, thereby further optimizing the program.
[0015] To sum up, after accessing the rehabilitation training program library, user group training samples are collected for any rehabilitation program, so as to obtain a comprehensive and accurate user sample data set, providing strong data support for subsequent utility analysis, program optimization and delivery.
[0016] The utility analysis module M300 is used to perform utility analysis on each renal rehabilitation program according to the user sample data set, and output a plurality of program failure indicators corresponding to the plurality of renal rehabilitation programs respectively.
[0017] When conducting a utility analysis of renal rehabilitation programs, the key is to use the acquired user sample data set to evaluate the effectiveness of each program, involving an in-depth analysis of changes in user rehabilitation data, rehabilitation recovery effects, and user training feedback. Furthermore, the utility analysis aims to reveal the effectiveness of each rehabilitation program in actual applications. By analyzing the user rehabilitation data change samples, we can understand the improvement in the patient's physical condition, physiological indicators, etc. after receiving different rehabilitation programs. Quantitative analysis is used to provide an intuitive evaluation basis, showing the advantages and disadvantages of each program in different aspects.
[0018] At the same time, user rehabilitation recovery effect samples are also an important reference for evaluating the effectiveness of the program. On the one hand, they take into account the patient's physiological recovery condition, and on the other hand, they involve the improvement of the patient's subjective feelings, such as symptom relief, improved quality of life, etc. These data often need to be obtained through patient self-evaluation, doctor evaluation or questionnaire surveys, in order to fully reflect the comprehensive effect of the rehabilitation program.
[0019] In addition, user training feedback samples cannot be ignored in utility analysis. Patient training feedback directly reflects the patient's body's acceptance, satisfaction with the rehabilitation program, and actual feelings during use; it helps understand the practicality and applicability of the program, thereby identifying existing problems and directions for improvement.
[0020] According to the set evaluation criteria, multiple failure indicators corresponding to multiple renal rehabilitation programs are output (such as too long rehabilitation time, poor recovery effect, low patient satisfaction), which can be used for quantitative scoring values or classification labels (such as effective, general, ineffective, etc.); through the failure indicators, the difference in the effects of various rehabilitation programs in actual applications can be intuitively seen, and user groups with poor performance can be identified, providing strong data support for subsequent classification, optimization and delivery.
[0021] The rehabilitation program division module M400 is used to divide the rehabilitation training program library according to the multiple program failure indicators to obtain a first type of renal rehabilitation program and a second type of renal rehabilitation program, wherein the first type of renal rehabilitation program is a set of programs whose program failure indicators are less than or equal to a preset failure threshold, and the second type of renal rehabilitation program is a set of programs whose program failure indicators are greater than the preset failure threshold.
[0022] After completing the utility analysis of each renal rehabilitation program and outputting the corresponding program failure indicators, the rehabilitation training program library is then divided according to the failure indicators. The purpose is to classify the programs according to their effects so that they can be subsequently optimized and delivered in a targeted manner. Furthermore, the division process follows the following principles: Preset failure threshold: A preset failure threshold needs to be set as the classification standard. The preset failure threshold is determined based on actual conditions and demand analysis, and aims to distinguish between rehabilitation programs with good effects and those with poor effects.
[0023] The rehabilitation training program library is divided according to the multiple program failure indicators: the first category of renal rehabilitation programs corresponds to a set of programs whose program failure indicators are less than or equal to a preset failure threshold, which are considered to be effective and worthy of promotion, and can significantly improve the patient's rehabilitation condition in practical applications; the second category of renal rehabilitation programs corresponds to a set of programs whose program failure indicators are greater than the preset failure threshold, which are considered to be ineffective and need further optimization, and have deficiencies in certain aspects, such as insignificant rehabilitation effects and low patient acceptance.
[0024] Through this classification method, we can clearly identify the high-quality programs and programs to be optimized in the rehabilitation training program library. For the first type of renal rehabilitation programs, they can be directly launched and promoted; while for the second type of renal rehabilitation programs, it is necessary to further analyze the reasons and take corresponding optimization measures to improve their effects.
[0025] In addition, the setting of the preset failure threshold needs to be carefully considered; if the threshold is set too high, a large number of solutions that should be optimized will be misjudged as high-quality solutions; if the threshold is set too low, some solutions with good actual results will be mistakenly classified as solutions to be optimized. Therefore, when setting the preset failure threshold, full reference should be made to historical data, expert opinions, actual needs and other factors.
[0026] The distillation learning training module M500 is used to take the preset failure threshold as the training target, perform distillation learning training on the second type of renal rehabilitation program based on the first type of renal rehabilitation program, and update the second type of renal rehabilitation program; the program delivery module M600 is used to deliver the first type of renal rehabilitation program and the updated second type of renal rehabilitation program.
[0027] After the programs in the rehabilitation training program library are divided into the first type of renal rehabilitation programs and the second type of renal rehabilitation programs according to the failure indicators, the key is to use the preset failure threshold as the training target to perform distillation learning training on the second type of renal rehabilitation programs, in order to improve the effect of the second type of renal rehabilitation programs and make it reach or approach the level of the first type of programs; distillation learning training is a knowledge transfer technology that allows the teacher model (i.e., the first type of renal rehabilitation program with good effects) to guide the student model (i.e., the second type of renal rehabilitation program to be optimized) to impart the knowledge and experience of the teacher model to the student model; in this process, the output of the teacher model is used as a goal or reference for student model training to guide the learning direction of the student model.
[0028] The specific training process includes: extracting key feature information from the first type of renal rehabilitation program. The extracted key feature information is the key to the good effect of the program. At the same time, corresponding features are extracted from the second type of renal rehabilitation program as the initial state of the student model. Based on the difference between the preset failure threshold and the failure index of the second type of renal rehabilitation program, a loss function is constructed. The loss function will be used to measure the difference between the output of the student model and the output of the teacher model, as well as the difference between the output of the student model and the preset target (i.e., the failure threshold); the parameters of the second type of renal rehabilitation program (i.e., the student model) are iteratively updated by minimizing the loss function. During the iterative training process, the student model will continuously try to imitate the output of the teacher model and strive to reduce the difference with the preset failure threshold; during the training process, the student model needs to be evaluated and verified regularly to ensure that its performance is constantly improving, which usually involves testing on a new user sample data set to check whether the rehabilitation effect of the student model has improved; when the failure index of the second type of renal rehabilitation program is reduced to below or close to the preset failure threshold after distillation learning training, it can be considered that these programs have been effectively optimized. At this time, the updated second type of renal rehabilitation program will be launched together with the first type of renal rehabilitation program for use by more patients.
[0029] Through distillation learning training, the performance of poorly effective programs in the rehabilitation training program library can be effectively improved so that they can better meet the rehabilitation needs of patients. At the same time, it provides an effective way for the continuous optimization and upgrading of rehabilitation programs.
[0030] Furthermore, if Figure 2 As shown, the distillation learning training module M500 is used to perform the following method:
[0031] Based on the first type of renal rehabilitation program, feature extraction is performed to obtain data for a first initialization model; based on the second type of renal rehabilitation program, feature extraction is performed to obtain data for a second initialization model; a loss function is introduced based on the difference between the preset failure threshold and the failure index of the corresponding program, and the data of the first initialization model is used to perform distillation loss calculation on the data of the second initialization model, and the data of the second initialization model is iteratively updated by minimizing the loss function to update the second type of renal rehabilitation program.
[0032] In the process of distillation learning training to optimize the second type of renal rehabilitation program, it is necessary to extract features from the first and second types of rehabilitation programs respectively, and construct corresponding initialization models. Furthermore, feature extraction and initialization model construction are performed. Specifically, key features are extracted from the first type of renal rehabilitation program, which are usually important factors leading to good program effects, including the specific steps, intensity, duration, and specific needs of the patient group of the rehabilitation method; the first initialization model is constructed using the extracted features. The first initialization model represents the core knowledge and experience of the first type of rehabilitation program and is the teacher model for subsequent distillation learning; features are extracted from the second type of renal rehabilitation program, including certain factors that lead to poor program effects, or potential points that the program itself has but has not been fully utilized; a second initialization model is constructed based on the extracted features. The second initialization model is a student model, and the goal is to imitate the teacher model (first initialization model) through distillation learning and improve its performance.
[0033] Loss function design and distillation loss calculation. Specifically, in order to measure the difference between the student model and the teacher model, and the difference between the student model output and the preset failure threshold, it is necessary to design a suitable loss function. The loss function consists of two parts: one part is used to measure the error between the predicted value and the true value (or target value) (such as mean square error), and the other part is used to measure the difference between the two models in the output probability distribution (such as KL divergence); the difference between the preset failure threshold and the failure index of the second type of renal rehabilitation program (ie, the student model) is used as part of the training objective and introduced into the loss function. Then, the data of the first initialized model (teacher model) is used to perform distillation loss calculation on the data of the second initialized model (student model), aiming to iteratively update the parameters of the student model by minimizing the loss function.
[0034] Iterative update and model optimization. Specifically, the parameters of the second initialized model are continuously updated iteratively by minimizing the loss function. In each iteration, the parameters of the student model are adjusted according to the output of the teacher model and the current output of the student model to make it closer to the performance level of the teacher model. As the iterations proceed, the performance of the student model will gradually improve and its failure index will gradually decrease. When the failure index of the student model drops below or approaches the preset failure threshold, the model can be considered to have been effectively optimized.
[0035] Update the second type of renal rehabilitation program. Specifically, after completing the distillation learning training, update the second type of renal rehabilitation program by comparing with the optimized second initialization model. These updated second type of renal rehabilitation programs will have better effects and be closer to the level of the first type of renal rehabilitation programs. The updated second type of renal rehabilitation programs will be put into use together with the first type of renal rehabilitation programs for more patients to use, thus ensuring that the programs in the rehabilitation training program library have higher effects and reliability.
[0036] Furthermore, the training program analysis system for renal rehabilitation is also used to perform the following method:
[0037] The expression of the loss function includes: ;
[0038] in, is the loss function, is the weight parameter, which controls the influence of the mean square error term on the total loss. is the weight parameter, which controls the influence of the KL divergence term on the total loss. For input sample The true label or target value, For all samples The mean square error loss is used to measure the second initialization model The difference between the predicted value and the true value, is the KL divergence, which measures the first initialized model and the second initialized model In the sample The difference between the above output probability distributions.
[0039] The expression of the loss function includes: , further, is a weight parameter that controls the influence of the mean square error term on the total loss, so as to balance the model's ability to fit the true label and its ability to imitate the output distribution of the teacher model (the first initialized model); That is, the mean square error term, which is the sum of all samples The mean square error loss of the second initialization model is calculated The predicted value of With target value The sum of squares of the differences between , used to measure the accuracy of the model prediction; is a weight parameter that controls the influence of the KL divergence term on the total loss. The KL divergence is a method to measure the difference between two probability distributions. By adjusting the value of β, the strength of the model in imitating the output distribution of the teacher model is controlled; is the KL divergence, which measures the first initialized model and the second initialized model In the sample The difference between the above output probability distributions, It is measured that when the second initialization model is used The output distribution of the first initialized model is approximated by The required information increment is used to encourage the second initialization model when the output distribution of Try to keep the output distribution as close to the first initialization model as possible , by adjusting the size of the weight, the accuracy of the second initialization model and the imitation ability of the first initialization model are balanced.
[0040] The entire loss function comprehensively considers the accuracy of model prediction (through the mean square error term) and the ability to imitate the output distribution of the teacher model (through the KL divergence term); by adjusting the values of α and β, these two aspects can be flexibly balanced to optimize the performance of the second initialized model (i.e., the second type of renal rehabilitation program); during the training process, by minimizing this loss function, the parameters of the second initialized model can be gradually updated so that its output is a rehabilitation program that is close to the first initialized model and has better effects.
[0041] Furthermore, the training program analysis system for renal rehabilitation is also used to perform the following method:
[0042] The calculation formula of KL divergence is as follows: ;
[0043] For the first initialized model pair sample The probability prediction output is Initialize the model for the second sample The probability prediction output is The logarithmic function is used to calculate information gain or loss.
[0044] KL divergence is also called relative entropy, which is used to measure the difference between two probability distributions. In the embodiment of the present application, it is used in distillation learning and other scenarios to compare the output distribution difference between the teacher model (first initialization model) and the student model (second initialization model), where: For the first initialized model pair sample The probability prediction output is a probability distribution, indicating that the first initialization model believes that the sample the probability of belonging to each possible category; Initialize the model for the second sample The probability prediction output is also a probability distribution. In the early stage of distillation learning, The corresponding distribution and There are certain differences in the corresponding distributions; The logarithmic function is used to calculate information gain or loss, helping to quantify the difference between two distributions; performing a sum operation means summing the KL divergence over all possible outputs (or categories), which usually means performing the above logarithmic calculation for each category of the model output and adding the results for all categories to obtain a single numerical value to measure the overall difference between the two distributions.
[0045] The KL divergence is non-negative and takes the value of 0 if and only if the two distributions are exactly the same; in the context of distillation learning, by minimizing the KL divergence, the second initialized model (student model) is encouraged to imitate the output distribution of the first initialized model (teacher model) as much as possible while maintaining computational efficiency, thereby improving its performance. This mechanism enables the student model to reduce computational efficiency without sacrificing too much accuracy and achieve more efficient reasoning.
[0046] Further, after the second type of renal rehabilitation program is updated, the updated second type of renal rehabilitation program is tested, and the training program analysis system for renal rehabilitation is used to perform the following method:
[0047] Collect an updated user sample data set; perform utility analysis on the updated user sample data set, and output an updated program failure index; if the updated program failure index is less than or equal to the preset failure threshold, activate a delivery instruction, and deliver the first category renal rehabilitation program and the updated second category renal rehabilitation program according to the delivery instruction.
[0048] After updating the second-category renal rehabilitation program, a series of verification steps are required to ensure its effectiveness and applicability. Furthermore, to ensure the representativeness and accuracy of the verification process, it is necessary to collect updated user sample data sets. These samples should cover kidney patients with different conditions, ages, genders and other background information to ensure the comprehensiveness of the verification. A detailed utility analysis of the newly collected user sample data sets is conducted to evaluate the actual effect of the updated second-category renal rehabilitation program, including analysis of changes in patients' rehabilitation data, rehabilitation recovery effects, and patients' training feedback on the new program.
[0049] According to the results of the utility analysis, the program failure index of the updated second-category renal rehabilitation program is calculated and compared with the preset failure threshold to evaluate the improvement effect of the program; the updated program failure index is compared with the preset failure threshold to determine whether the preset conditions are met. If the updated program failure index is less than or equal to the preset failure threshold, it means that the updated second-category renal rehabilitation program has achieved the expected improvement effect; if the updated program failure index meets the preset conditions, that is, it is less than or equal to the preset failure threshold, the delivery instruction is activated, indicating that the updated second-category renal rehabilitation program has passed strict inspection and is ready to be put into actual use; according to the delivery instruction, the first-category renal rehabilitation program and the updated second-category renal rehabilitation program are delivered, which means that the first-category renal rehabilitation program and the updated second-category renal rehabilitation program will be pushed to relevant medical institutions, rehabilitation centers or individual patients for use in the actual rehabilitation process.
[0050] It should be noted that during the entire inspection process, the authenticity and accuracy of the data should be ensured to avoid misjudgment due to data errors; the preset failure threshold should be reasonably set according to the actual situation to ensure that it can reflect the effectiveness of the program while avoiding overly harsh evaluation criteria; sufficient communication and coordination should be carried out before delivery to ensure that the recipient understands the specific content and usage of the program. Ensure the effectiveness and reliability of the updated second-category renal rehabilitation program in actual application, so as to provide better rehabilitation services for kidney patients.
[0051] Furthermore, the training program analysis system for renal rehabilitation is used to perform the following method:
[0052] A user group training sample is collected for any rehabilitation program in the rehabilitation training program library, wherein the user group includes a user group quantity constraint index, a user group medical record representativeness, a user group inter-class difference constraint index, and a user group data continuity.
[0053] When collecting user group training samples for any rehabilitation program in the rehabilitation training program library, in order to ensure that the collected samples are representative, comprehensive and continuous, further, the user group quantity constraint index is a quantitative index (for example, at least 100 user groups), which is used to stipulate the lower limit of the number of samples in each user group. A sufficient number of samples can ensure the stability and reliability of the statistical results. The setting of the quantity constraint index should be based on statistical principles and actual needs, and it is necessary to avoid the deviation caused by too small a sample size and consider the feasibility of actual operation. The representativeness of the user group medical records means that the collected user groups should be able to cover a variety of kidney patients with different conditions, courses of disease, ages, genders, physical constitutions, etc., to ensure that the user samples can represent the diversity of kidney patients, thereby reflecting the general characteristics and individual differences of kidney patients, and improving the generalization ability and applicability of rehabilitation programs. In actual operations, user groups can be constructed by screening typical cases and representative patients. The inter-class difference constraint index of the user group is the degree of difference between different categories within the user group, including multiple kidney case categories, but the difference between the categories is small, involving different types of kidney diseases, different rehabilitation stages or different treatment responses, etc.; by setting the inter-class difference constraint index, it can be ensured that the collected samples have a certain degree of distinction between different categories, which is helpful to train a more accurate and personalized rehabilitation plan. The user group data continuity refers to the user group's ability to participate in the research for a long time and provide continuous data support. In the process of analyzing and optimizing the rehabilitation training plan, it is necessary to continuously collect the patient's rehabilitation data and feedback information. Therefore, the user group should be able to maintain a certain stability and continuity to obtain complete rehabilitation data and long-term effects. In order to achieve data continuity, regular follow-up, establishment of a patient database and formulation of a long-term research plan can be adopted.
[0054] In summary, when collecting user group training samples for any rehabilitation program in the rehabilitation training program library, key factors such as user group quantity constraint indicators, user group medical record representativeness, user group inter-class difference constraint indicators, and user group data continuity should be considered comprehensively to ensure that the collected training samples are of high quality and representativeness, providing a solid data foundation for subsequent utility analysis and program optimization.
[0055] Further, after obtaining the first type of renal rehabilitation program and the updated second type of renal rehabilitation program, the training program analysis system for renal rehabilitation is further used to perform the following method:
[0056] Perform feature identification on the first type of renal rehabilitation program and the updated second type of renal rehabilitation program to obtain a program feature set; use a recursive feature elimination algorithm to screen features whose impact on rehabilitation effect is greater than a preset impact from the program feature set, and output a valid feature set; define a third initialization model, use the valid feature set as a new training data set to train the third initialization model, and output a converged third initialization model; optimize the second type of renal rehabilitation program based on the converged third initialization model.
[0057] After obtaining the first-category renal rehabilitation program and the updated second-category renal rehabilitation program, in order to further improve the effectiveness and accuracy of the rehabilitation program, detailed feature identification was performed on the first-category renal rehabilitation program and the updated second-category renal rehabilitation program, including the specific content of the rehabilitation program, implementation steps, technology used, targeted disease types, patient age and gender distribution, rehabilitation cycle, recovery effect indicators and other dimensions; all identified features were organized into a program feature set to provide a basis for subsequent feature screening and optimization.
[0058] Use the recursive feature elimination algorithm to screen out effective features. The recursive feature elimination (RFE) algorithm can screen out features whose impact on the rehabilitation effect is greater than the preset impact from the updated second-category renal rehabilitation program. The RFE algorithm works by building a model and iteratively considering fewer and fewer features (eliminating the weakest feature in each iteration) until the required number of features is reached or the model performance is no longer significantly improved. In this process, a preset impact threshold can be set. Only when the impact of a feature on the rehabilitation effect exceeds the preset impact threshold, it is considered a valid feature and retained. The output is a valid feature set containing valid features.
[0059] Based on the screened effective feature set, a new model is defined as the third initialization model, and the third initialization model is used to further analyze and optimize the rehabilitation plan; the type of the third initialization model can be selected according to actual conditions, such as a machine learning model, a deep learning model or a traditional statistical model; the third initialization model is trained using the effective feature set as a new training data set; during the training process, the parameters and structure of the model are continuously adjusted to improve the prediction accuracy and generalization ability of the model; a converged third initialization model is obtained, and the third initialization model can accurately reflect the relationship between the effective features of the rehabilitation plan and the rehabilitation effect.
[0060] The converged third initialization model is used to further analyze and optimize the second type of renal rehabilitation program. By analyzing the relationship between the predicted results output by the model and the rehabilitation effect, the problems and shortcomings in the second type of rehabilitation program are found out. In response to these problems and shortcomings, the second type of rehabilitation program is modified and improved in a targeted manner, for example, the order of rehabilitation steps is adjusted, the application method of certain technologies is changed, certain rehabilitation content is increased or decreased, etc.; an optimized and more effective second type of renal rehabilitation program is obtained. The optimized second type of renal rehabilitation program will better meet the rehabilitation needs of patients and improve the rehabilitation effect.
[0061] Furthermore, the training solution library access module M100 is also used to execute the following method:
[0062] Determine whether the program quantification index of the rehabilitation training program library is greater than the preset quantification index; if it is greater than the preset quantification index, divide the rehabilitation training program library to obtain a first type of renal rehabilitation program, a second type of renal rehabilitation program and a third type of renal rehabilitation program; wherein, the first type of renal rehabilitation program is a set of programs whose program failure index is less than or equal to a first preset failure threshold, the second type of renal rehabilitation program is a set of programs whose program failure index is greater than the first preset failure threshold and less than the second preset failure threshold, and the third type of renal rehabilitation program is a set of programs whose program failure index is greater than or equal to the second preset failure threshold; perform distillation learning on the third type of renal rehabilitation program according to the second type of renal rehabilitation program, and perform distillation learning on the second type of renal rehabilitation program according to the first type of renal rehabilitation program.
[0063] When processing the rehabilitation training program library, in order to ensure that the classification of the programs is reasonable and effective, it is necessary to set a program quantitative index and compare it with the preset quantitative index. Specifically, the quantitative index of each program in the rehabilitation training program library is calculated (based on a comprehensive evaluation of multiple factors, such as rehabilitation effect, patient satisfaction, rehabilitation cycle, etc.); the calculated program quantitative index is compared with the preset quantitative index; if the quantitative index of a certain program is greater than the preset quantitative index, the program library is further divided into more detailed groups. The division results include three types of renal rehabilitation programs: the first type of renal rehabilitation program: a set of programs with a program failure index less than or equal to the first preset failure threshold, which usually show higher rehabilitation effects and patient satisfaction; the second type of renal rehabilitation program: a set of programs with a program failure index greater than the first preset failure threshold and less than the second preset failure threshold, which is slightly inferior to the first type of program in terms of rehabilitation effect, but still has certain value; the third type of renal rehabilitation program: a set of programs with a program failure index greater than or equal to the second preset failure threshold, which performs poorly in terms of rehabilitation effect and needs further improvement or elimination.
[0064] The distillation learning technology is used to optimize the divided set of schemes. Distillation learning is a knowledge transfer technology that can transfer the knowledge of one model (teacher model) to another model (student model). In the embodiment of the present application, the second type of renal rehabilitation scheme can be used as a student model, and the first type of renal rehabilitation scheme can be used as a teacher model. Through distillation learning, the second type of scheme can learn the advantages of the first type of scheme, thereby improving its rehabilitation effect. At the same time, the second type of renal rehabilitation scheme can be considered to be used for distillation learning of the third type of renal rehabilitation scheme, which will help improve the performance of the third type of scheme to a certain extent.
[0065] It should be noted that the setting of quantitative indicators should be based on sufficient data analysis and expert opinions to ensure that they can objectively and comprehensively reflect the effect of the rehabilitation program; the determination of the preset failure threshold should be based on clinical experience and actual needs, considering both the improvement of rehabilitation effects and avoiding overly harsh evaluation criteria that lead to excessive elimination of programs; the implementation of distillation learning requires the selection of appropriate algorithms and parameters, and sufficient experimental verification to ensure its effectiveness and stability. Through the above process, the rehabilitation training program library is scientifically and rationally divided and optimized to provide more accurate and effective rehabilitation services for kidney patients.
[0066] In summary, the beneficial effects of the embodiments of the present application are:
[0067] By analyzing a large amount of user sample data, the utility of each renal rehabilitation program is analyzed and program failure indicators are output, so as to scientifically evaluate the effectiveness of each program and improve the quality of the overall rehabilitation program.
[0068] Rehabilitation plans are divided according to the failure indicators of the plans. Rehabilitation plans that are more in line with individual characteristics are customized according to the characteristics and needs of different user groups. Plans with poor results are improved through distillation learning training, thereby speeding up training.
[0069] Through continuous monitoring and evaluation, the rehabilitation plan is constantly adjusted to make it more in line with the patient's actual needs and condition, and more adaptable to changes in patient needs.
[0070] Since the method is used to judge whether the quantitative index of the rehabilitation training program library is greater than the preset quantitative index; if it is greater than the preset quantitative index, the rehabilitation training program library is divided to obtain the first type of renal rehabilitation program, the second type of renal rehabilitation program and the third type of renal rehabilitation program; among them, the first type of renal rehabilitation program is a set of programs whose program failure index is less than or equal to the first preset failure threshold, the second type of renal rehabilitation program is a set of programs whose program failure index is greater than the first preset failure threshold and less than the second preset failure threshold, and the third type of renal rehabilitation program is a set of programs whose program failure index is greater than or equal to the second preset failure threshold; the third type of renal rehabilitation program is distilled and learned according to the second type of renal rehabilitation program, and the second type of renal rehabilitation program is distilled and learned according to the first type of renal rehabilitation program. Through the above process, the rehabilitation training program library is scientifically and reasonably divided and optimized, providing more accurate and effective rehabilitation services for kidney patients.
[0071] In summary, any step can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor, and no unnecessary restrictions are made here.
[0072] Furthermore, the above-mentioned technical scheme only reflects the preferred technical scheme of the technical scheme of the embodiment of the present application. Some changes that may be made to some parts thereof by technicians in this technical field all reflect the novel principles of the embodiment of the present application. Obviously, technicians in this field can make various changes and modifications to the present application without departing from the scope of the present application.
Claims
1. A training program analysis system for renal rehabilitation, characterized in that: include: A training program library access module, used to access a rehabilitation training program library, wherein the rehabilitation training program library stores a plurality of renal rehabilitation programs; A training sample collection module, used to collect user group training samples for any rehabilitation program in the rehabilitation training program library to obtain a user sample data set, wherein the user sample data set includes user rehabilitation data change samples, user rehabilitation recovery effect samples and user training feedback samples; A utility analysis module, configured to perform utility analysis on each renal rehabilitation program according to the user sample data set, and output a plurality of program failure indicators corresponding to the plurality of renal rehabilitation programs respectively; A rehabilitation program classification module, used for classifying the rehabilitation training program library according to the multiple program failure indicators to obtain a first type of renal rehabilitation program and a second type of renal rehabilitation program, wherein the first type of renal rehabilitation program is a set of programs whose program failure indicators are less than or equal to a preset failure threshold, and the second type of renal rehabilitation program is a set of programs whose program failure indicators are greater than the preset failure threshold; a distillation learning training module, configured to perform distillation learning training on the second type of renal rehabilitation program based on the first type of renal rehabilitation program, taking the preset failure threshold as a training target, and updating the second type of renal rehabilitation program; A program delivery module, used for delivering the first type of renal rehabilitation program and the updated second type of renal rehabilitation program; The distillation learning training module includes: extracting features based on the first type of kidney rehabilitation program to obtain data of a first initialization model; Perform feature extraction based on the second type of renal rehabilitation program to obtain data of a second initialization model; A loss function is introduced by using the difference between the preset failure threshold and the failure index of the corresponding scheme, and the data of the second initialization model is used to perform distillation loss calculation on the data of the second initialization model, and the data of the second initialization model is iteratively updated by minimizing the loss function to update the second type of renal rehabilitation scheme; The expression of the loss function includes: ; in, is the loss function, is the weight parameter, which controls the influence of the mean square error term on the total loss. is the weight parameter, which controls the influence of the KL divergence term on the total loss. For input sample The true label or target value, For all samples The mean square error loss is used to measure the second initialization model The difference between the predicted value and the true value, is the KL divergence, which measures the first initialized model and the second initialized model In the sample The difference between the above output probability distributions; The calculation formula of KL divergence is as follows: ; For the first initialized model pair sample The probability prediction output is Initialize the model for the second sample The probability prediction output is It is a logarithmic function used to calculate information gain or loss; Collecting user group training samples for any rehabilitation program in the rehabilitation training program library, wherein the user group includes a user group quantity constraint index, a user group medical record representativeness, a user group inter-class difference constraint index, and a user group data continuity; With the first category renal rehabilitation program and the updated second category renal rehabilitation program, it also includes: Performing feature recognition on the first type of renal rehabilitation program and the updated second type of renal rehabilitation program to obtain a program feature set; Using a recursive feature elimination algorithm to select features whose impact on the rehabilitation effect is greater than a preset impact from the feature set of the scheme, and outputting a valid feature set; defining a third initialization model, training the third initialization model using the effective feature set as a new training data set, and outputting a converged third initialization model; optimizing the second type of renal rehabilitation program according to the converged third initialized model; The training program library access module also includes: Determining whether the program quantification index of the rehabilitation training program library is greater than a preset quantification index; If it is greater than the preset quantitative index, the rehabilitation training program library is divided to obtain a first-category renal rehabilitation program, a second-category renal rehabilitation program, and a third-category renal rehabilitation program; The first type of renal rehabilitation program is a set of programs whose program failure index is less than or equal to a first preset failure threshold, the second type of renal rehabilitation program is a set of programs whose program failure index is greater than the first preset failure threshold and less than a second preset failure threshold, and the third type of renal rehabilitation program is a set of programs whose program failure index is greater than or equal to the second preset failure threshold; The third type of renal rehabilitation program is distilled and learned according to the second type of renal rehabilitation program, and the second type of renal rehabilitation program is distilled and learned according to the first type of renal rehabilitation program.
2. The training program analysis system for renal rehabilitation according to claim 1, characterized in that: After the second category renal rehabilitation program is updated, the updated second category renal rehabilitation program is tested, including; Collect updated user sample data sets; Perform utility analysis on the updated user sample data set and output the updated solution failure index; If the updated program failure index is less than or equal to the preset failure threshold, the delivery instruction is activated, and according to the delivery instruction, the first type of renal rehabilitation program and the updated second type of renal rehabilitation program are delivered.
