Intelligent diagnosis and treatment method and system for knee osteoarthritis based on machine learning technology

Through the intelligent diagnosis and treatment system of knee osteoarthritis based on machine learning, the treatment plan is optimized using deep reinforcement learning and reward function, the problem of insufficient personalized recommendations in knee osteoarthritis treatment is solved, the optimization and efficiency of treatment plan are achieved, and the reduction of the treatment plan is achieved. The economic and time cost of the patient.

CN118737365BActive Publication Date: 2024-11-05WEST CHINA HOSPITAL SICHUAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411219038.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-02
Publication Date
2024-11-05
Estimated Expiration
2044-09-02

AI Technical Summary

Technical Problem

现有技术在膝骨关节炎的诊疗中难以提供个性化的治疗推荐,传统治疗方案无法满足患者的多样化需求,且长期治疗带来的时间和经济成本沉重,复诊及随访难以达到理想效果。

Method used

Adopting an intelligent diagnosis and treatment system for knee osteoarthritis based on machine learning technology, through data preprocessing, feature extraction, feature screening, rehabilitation treatment effect simulation and decision-making modules, combined with deep reinforcement learning and reward functions, optimize treatment plan recommendations and provide optimal treatment plan.

Benefits of technology

The simulation and optimization of multiple treatment plans have been achieved, the treatment efficiency has been improved, the resource waste has been reduced, the treatment has been improved, and the economic and time cost of patients has been reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118737365B_ABST
    Figure CN118737365B_ABST
Patent Text Reader

Abstract

The present invention discloses an intelligent diagnosis and treatment method and system for knee osteoarthritis based on machine learning technology: a data preprocessing module, used to obtain a KOA database; a feature extraction module, used to obtain multiple feature quantities in the KOA database; a KOA feature screening module, which extracts a key feature set from multiple feature quantities; a KOA rehabilitation treatment effect simulation module, which simulates the state of the patient at the next time point after adopting the initial treatment plan according to the key feature set; a KOA rehabilitation treatment decision module, which recommends the treatment plan that the patient should choose within a preset time through preset decision rules; a KOA intelligent treatment auxiliary decision module, which optimizes the KOA rehabilitation treatment decision module by learning the optimal strategy function to obtain the optimal treatment plan; the beneficial effects achieved by the present invention are: it realizes the simulation of corresponding treatment results of multiple possible treatment plans, optimizes the treatment plan, and solves the problems of KOA patients in data analysis and personalized treatment recommendation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent diagnosis and treatment of arthritis, and in particular to an intelligent diagnosis and treatment method and system for knee osteoarthritis based on machine learning technology. Background Art

[0002] Knee osteoarthritis (KOA) is a highly disabling disease. With the aging of the global population, the prevalence of KOA has shown a significant upward trend. First, due to the large number of patients and insufficient rehabilitation service capacity, traditional offline treatment programs cannot meet the rehabilitation needs of patients one-to-one; second, KOA requires long-term treatment. The time cost of long-term treatment, the transportation costs to and from the hospital, and the rehabilitation costs bring a heavy burden to the family and society. The re-examination and follow-up of KOA patients has become a greater problem, and it is often difficult to achieve the ideal treatment effect;

[0003] Although artificial intelligence technologies, especially deep learning, machine vision and natural language processing, have been partially applied in the diagnosis, treatment and disease management of KOA, unfortunately, these studies mainly focus on static data analysis and preliminary prediction. The progress made in these two aspects cannot cover up the obvious shortcomings in intelligent and personalized treatment recommendations.

[0004] In view of the deficiencies of the prior art, the present invention proposes an intelligent diagnosis and treatment method and system for knee osteoarthritis based on machine learning technology. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide an optimal treatment plan based on the patient's time series data. The present invention proposes an intelligent diagnosis and treatment method and system for knee osteoarthritis based on machine learning technology, which realizes the simulation of corresponding treatment results of multiple possible treatment plans, optimizes the treatment plan, and solves the problems in medical data analysis and personalized treatment recommendations for KOA patients.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] The intelligent diagnosis and treatment system for knee osteoarthritis based on machine learning technology includes:

[0008] A data preprocessing module is used to obtain a KOA database based on multi-dimensional scale data of KOA patients;

[0009] A feature extraction module is used to obtain multiple feature quantities in the KOA database;

[0010] KOA feature screening module, used to extract key feature sets from multiple feature quantities using an iterative feature elimination strategy;

[0011] The KOA rehabilitation treatment effect simulation module is used to simulate the state of the patient at the next time point after the initial treatment plan is adopted according to the key feature set; the state at the next time point is recorded under the corresponding treatment plan of the corresponding patient to obtain the updated KOA database;

[0012] The KOA rehabilitation treatment decision module is used to recommend the treatment plan that the patient should choose within a preset time according to the updated KOA database through preset decision rules.

[0013] Furthermore, the intelligent diagnosis and treatment system for knee osteoarthritis also includes: a KOA intelligent treatment auxiliary decision module, which is used to optimize the KOA rehabilitation treatment decision module by learning the optimal strategy function to obtain the optimal treatment plan;

[0014] The KOA intelligent treatment auxiliary decision module includes a reward submodule, a deep reinforcement submodule and a decision evaluation submodule;

[0015] The reward submodule is used to evaluate the effectiveness of the treatment plan through a reward function to obtain an immediate feedback value;

[0016] The deep reinforcement submodule is used to learn the optimal strategy function according to the immediate feedback value to obtain the cumulative reward value of each treatment plan;

[0017] The decision evaluation submodule is used to output the optimal treatment plan according to the accumulated reward value; the optimal treatment plan is the treatment plan combination with the highest cumulative reward value combination score after the treatment plans are combined.

[0018] Furthermore, the expression of the reward function is as follows:

[0019]

[0020] Where:

[0021] Indicates the immediate feedback value;

[0022] : KOA pain score at week t;

[0023] : KOA pain score at week t+1;

[0024] : The first threshold of KOA pain score;

[0025] : The second threshold of KOA pain score;

[0026] is a constant.

[0027] Furthermore, the reward function includes three cases;

[0028] when < To improve the situation, the reward function is calculated as:

[0029] ;

[0030] When the KOA pain score decreases, it means that the selected treatment plan shows an improvement in the patient's condition. At this time, it is necessary to provide a reward for any degree of decrease in the immediate feedback value, rewarding the KOA pain score to decrease from a higher score, so as to obtain a greater decrease in the KOA pain score;

[0031] when = When , it is a stable situation, then compare and thresholds for pain scores;

[0032] like > ,but ; High but stable KOA pain scores need to be penalized through immediate feedback values;

[0033] like ≦ ,but Very low but stable KOA pain scores need to be rewarded with immediate feedback values;

[0034] like < ≦ ,but ; Moderate but stable KOA pain scores were neither rewarded nor penalized through immediate feedback values;

[0035] when > When , it is a deteriorating situation, then the calculation formula of the reward function is;

[0036] ,

[0037] When the KOA pain score increases, it indicates that the situation is worsening and a penalty should be imposed on the increase in the KOA pain score, with a more severe penalty imposed on a relatively high score increase.

[0038] Furthermore, the data preprocessing module includes: a data cleaning submodule, a missing value processing submodule, a category imbalance processing submodule and a normalization submodule;

[0039] The data cleaning submodule is used to remove abnormal values ​​in the scale data to obtain normal value scale data;

[0040] The missing value processing submodule is used to fill in the independent missing values ​​in the normal value scale data to obtain the whole scale data;

[0041] The class imbalance processing submodule is used to obtain balanced scale data by resampling and cost sensitivity analysis of the whole scale data;

[0042] The normalization submodule is used to process the balance scale data by data normalization to obtain the KOA database.

[0043] Further, the KOA feature screening module includes: an initial random forest submodule and a random forest submodule;

[0044] The initial random forest submodule is used to obtain a random forest submodule through KOA database training;

[0045] The random forest submodule is used to adopt an iterative feature elimination strategy to screen multiple feature quantities in the KOA database to obtain a key feature set.

[0046] Furthermore, the KOA rehabilitation treatment effect simulation module includes: a sequence submodule, a history information encoding submodule, a multi-layer perception submodule and an update submodule.

[0047] The sequence submodule is used to obtain time series disease data in the key feature set; the time series disease data includes the patient's historical status and treatment plan;

[0048] The historical information encoding submodule is used to encode the historical status and treatment plan to obtain the historical information code;

[0049] The multi-layer perception submodule is used to predict the patient's state at the next time point based on the historical information encoding;

[0050] The updating submodule is used to record the status at the next time point into the corresponding treatment plan of the corresponding patient to update the KOA database.

[0051] Furthermore, the KOA rehabilitation treatment decision module includes: a four-tuple data submodule and a dynamic recommendation submodule.

[0052] The four-tuple data submodule is used to obtain the historical status, treatment plan, cumulative reward value and status at the next time point of each patient from the updated KOA database to form four-tuple data;

[0053] The dynamic recommendation submodule is used to predict the treatment plan for a preset time based on the four-tuple data, and recommend the treatment plan that the patient should choose within the preset time through a preset decision rule.

[0054] The present invention also provides an intelligent diagnosis and treatment method for knee osteoarthritis based on machine learning technology, comprising:

[0055] S001. Obtain the KOA database based on the multidimensional scale data of KOA patients;

[0056] S002, obtaining multiple feature quantities from the KOA database;

[0057] S003, extracting a key feature set from multiple feature quantities by adopting an iterative feature elimination strategy;

[0058] S004. Simulate the state of the patient at the next time point after the initial treatment plan is adopted according to the key feature set; record the state at the next time point under the corresponding treatment plan of the corresponding patient to update the KOA database;

[0059] S005. Based on the updated KOA database, recommend the treatment plan that the patient should choose within the preset time through the preset decision rules;

[0060] S006. Optimize the treatment plan by learning the optimal strategy function to obtain the optimal treatment plan.

[0061] Furthermore, extracting a key feature set from multiple feature quantities by adopting an iterative feature elimination strategy includes: constructing and training a random forest model, and based on the random forest model, adopting an iterative feature elimination strategy to screen multiple feature quantities in the KOA database to obtain a key feature set;

[0062] The random forest model is constructed and trained, including:

[0063] S301, constructing an initial random forest model;

[0064] S302, using self-service sampling, randomly selecting m features from multiple features in the KOA database to form a first training set;

[0065] S303, calculating the importance score of each feature in the first training set using out-of-bag error estimation;

[0066] S304, sorting all features in descending order according to importance scores;

[0067] S305, using step-by-step back-off search to iteratively remove the feature with the lowest importance score in the first training set;

[0068] S306, calculating the classification accuracy of the random forest model after removing a feature each time;

[0069] S307. When the classification accuracy is stabilized at a preset value, stop removing features to obtain a key feature set; the key feature set is a set of remaining features after removing some features in the first training set according to S305 to S307.

[0070] The present invention has the following advantages:

[0071] The present invention constructs a KOA rehabilitation treatment effect simulation module, which can simulate the patient's state change at the next time point under hundreds of possible treatment plans based on the patient's state, so that the results of different plans after implementation on the patient can be obtained without actually implementing the treatment plan, thereby optimizing the treatment strategy; this reduces unnecessary waste of resources and time, and greatly improves the doctor's treatment efficiency;

[0072] The present invention establishes a reward function in the KOA patient intelligent treatment auxiliary decision-making module, calculates the adaptive changes of the reward function for different situations, and constructs a deep reinforcement submodule on this basis. By learning the optimal strategy function, the treatment plan output by the KOA rehabilitation treatment effect simulation module is combined and optimized to obtain the optimal treatment plan;

[0073] The present invention extracts key feature sets from multiple feature quantities by adopting an iterative feature elimination strategy. Processing only the key feature sets is beneficial to improving the accuracy of simulating the state of the patient at the next time point after adopting the treatment plan, and reducing unnecessary calculations to achieve the effect of a lightweight system. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a system schematic diagram of the present invention;

[0075] Figure 2 It is a diagram of the method steps of the present invention;

[0076] Figure 3 A diagram of the steps to build and train a random forest model. DETAILED DESCRIPTION

[0077] The present invention is further described below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following description.

[0078] It should be noted that the directions or positional relationships indicated by "left", "right", etc. are based on the directions or positional relationships shown in the accompanying drawings, or are the directions or positional relationships in which the inventive product is usually placed when in use, or are the directions or positional relationships commonly understood by those skilled in the art. Such terms are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0079] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features and technical solutions in the embodiments may be combined with each other.

[0080] First aspect

[0081] See also Figure 1 The present invention provides an intelligent diagnosis and treatment system for knee osteoarthritis based on machine learning technology, comprising:

[0082] A data preprocessing module is used to obtain a KOA database based on the multi-dimensional scale data of KOA patients;

[0083] A feature extraction module is used to obtain multiple feature quantities in the KOA database;

[0084] KOA feature screening module, used to extract key feature sets from multiple feature quantities using an iterative feature elimination strategy;

[0085] The KOA rehabilitation treatment effect simulation module is used to simulate the patient's state at the next time point after the initial treatment plan is adopted according to the key feature set; the state at the next time point is recorded under the corresponding treatment plan of the corresponding patient to update the KOA database;

[0086] The KOA rehabilitation treatment decision module is used to recommend the treatment plan that the patient should choose within the preset time according to the updated KOA database through the preset decision rules;

[0087] The KOA intelligent treatment auxiliary decision module is used to optimize the KOA rehabilitation treatment decision module by learning the optimal strategy function to obtain the optimal treatment plan; the optimal treatment plan is the treatment plan combination with the highest cumulative reward value combination score after the treatment plan combination.

[0088] Before work, we are authorized to collect clinical data such as basic information, pain level, stiffness level, daily life, dynamic balance, quality of life, anxiety and depression of KOA patients, and convert these multi-dimensional clinical data into scale data; then obtain the KOA database through the data preprocessing module, and the KOA database provides basic data for the construction of subsequent modules.

[0089] The feature extraction module is used to obtain multiple feature quantities in the KOA database; then the KOA feature screening module is used to screen the feature quantities, and the features with the greatest impact on treatment decisions are screened from 122 features to obtain the key feature set; through feature screening, not only the accuracy of the simulation effect prediction after the construction of the KOA rehabilitation treatment effect simulation module can be improved, but also the questionnaire design can be optimized for the future collection of basic data of KOA patients, reducing the patient rehabilitation assessment time and improving the efficiency and accuracy of data collection.

[0090] Based on the screened key feature set, the KOA rehabilitation treatment effect simulation module was constructed through a recurrent neural network (RNN) to simulate the impact of different treatment plans on patients. And through sensitivity analysis, the prediction accuracy of the KOA rehabilitation treatment effect simulation module in different time windows was evaluated to ensure the stability and reliability of the KOA rehabilitation treatment effect simulation module.

[0091] The KOA intelligent treatment decision-making support module is constructed based on the deep reinforcement learning algorithm, and a reward function is set at the same time. Through a large amount of clinical data in the scale data, the effectiveness of the treatment plan combination decided by the KOA intelligent treatment decision-making support module in KOA treatment is verified, so as to obtain the optimal treatment plan.

[0092] Furthermore, the KOA intelligent treatment decision-making assistance module includes a reward submodule, a deep reinforcement submodule, and a decision-making evaluation submodule;

[0093] The reward submodule is used to evaluate the effectiveness of the treatment plan through the reward function to obtain immediate feedback value;

[0094] The expression of the reward function is as follows:

[0095]

[0096] Where:

[0097] : Instant feedback value;

[0098] : KOA pain score at week t;

[0099] : KOA pain score at week t+1;

[0100] : The first threshold of KOA pain score;

[0101] : The second threshold of KOA pain score;

[0102] is a constant.

[0103] In some specific embodiments, =0; =3; =0.1; =0.1; =0; =0.1.

[0104] In the calculation of the reward function, the calculation of the reward function includes three cases;

[0105] when < To improve the situation, the reward function is calculated as:

[0106] ;

[0107] When the KOA pain score decreases, it means that the selected treatment plan shows an improvement in the patient's condition. At this time, it is necessary to provide a reward for any degree of decrease in the immediate feedback value, rewarding the KOA pain score to decrease from a higher score, so as to obtain a greater decrease in the KOA pain score;

[0108] when = When , it is a stable situation, then compare Thresholds with KOA pain scores;

[0109] like > ,but ; High but stable KOA pain scores need to be penalized through immediate feedback values;

[0110] like ≦ ,but Very low but stable KOA pain scores need to be rewarded with immediate feedback values;

[0111] like < ≦ ,but ; Moderate but stable KOA pain scores were neither rewarded nor penalized through immediate feedback values;

[0112] when > When , it is a deteriorating situation, then the calculation formula of the reward function is;

[0113] ;

[0114] When the KOA pain score increases, it indicates that the situation is now deteriorating, and a penalty should be imposed on the increase in the KOA pain score, with a more severe penalty imposed on a relatively high score increase;

[0115] The reward function is adaptively calculated for different situations to obtain different immediate feedback values ​​suitable for complex transformation situations, so that the rewards and penalties applied through the immediate feedback values ​​are more accurate, so as to better adjust the subsequent solution selection.

[0116] The KOA intelligent treatment auxiliary decision module also includes: a deep reinforcement submodule, which is used to learn the optimal strategy function according to the immediate feedback value to obtain the cumulative reward value of each treatment plan; a decision evaluation submodule, which is used to output the optimal treatment plan according to the cumulative reward value;

[0117] The KOA intelligent treatment decision-making support module can intelligently optimize treatment plans based on the patient's dynamic state, improve treatment personalization and effectiveness, and reduce patients' medical time costs and financial burden.

[0118] Furthermore, the data preprocessing module includes: a data cleaning submodule, a missing value processing submodule, a category imbalance processing submodule and a normalization submodule;

[0119] The data cleaning submodule is used to remove abnormal values ​​in the scale data to obtain normal value scale data;

[0120] The missing value processing submodule is used to fill in the independent missing values ​​in the normal value scale data to obtain the whole scale data;

[0121] The class imbalance processing submodule is used to obtain balanced scale data by resampling and cost sensitivity analysis of the whole scale data;

[0122] The normalization submodule is used to process the balance scale data by data normalization to obtain the KOA database.

[0123] Furthermore, the KOA feature screening module includes: an initial random forest submodule and a random forest submodule;

[0124] The initial random forest submodule is used to obtain a random forest submodule through KOA database training;

[0125] The random forest submodule is used to select multiple feature quantities in the KOA database by adopting an iterative feature elimination strategy to obtain a key feature set;

[0126] The iterative feature elimination strategy is used to extract key feature sets from multiple feature quantities. Processing only the key feature sets is beneficial to improving the accuracy of simulating the patient's state at the next time point after the treatment plan is adopted, and reducing unnecessary calculations to achieve the effect of a lightweight system.

[0127] Furthermore, the KOA rehabilitation treatment effect simulation module includes: a sequence submodule, a history information encoding submodule, a multi-layer perception submodule and an update submodule;

[0128] The sequence submodule is used to obtain time series disease data in the key feature set; the time series disease data includes the patient's historical state and treatment plan; the sequence submodule includes: a state transition process function, the state transition process function represents the state transition process of the patient from the historical state at the historical time point t to the state at the next time point t+1, and the state transition process function expression is:

[0129]

[0130] Where:

[0131] : The patient's status at the next time point t+1;

[0132] : The historical status of the patient at time point t;

[0133] : Treatment options;

[0134] The historical information encoding submodule is used to encode the historical status and treatment plan to obtain the historical information code; the specific encoding process in the historical information encoding submodule includes: using a coding function to encode the historical status and treatment plan, and the expression of the coding function is:

[0135]

[0136] Where:

[0137] : historical information;

[0138] : Time point To the time point The sequence of patient states;

[0139] : Time point To the time point The sequence of treatment options;

[0140] : From time point t to historical time window In some embodiments, Set to 1 week;

[0141] The multi-layer perception submodule is used to predict the patient's state at the next time point based on the historical information encoding; the objective function of the multi-layer perception submodule is:

[0142]

[0143] Where:

[0144] : The i-th patient at time point The true value of the d-th state,

[0145] : The i-th patient at time point The predicted value of the d-th state;

[0146] N: total number of patients;

[0147] Ti: total number of time points t for the ith patient;

[0148] D: total number of states;

[0149] By reducing the difference between the predicted value, i.e. the patient's state at the next time point, and the actual value through the objective function, the patient's state at the next time point can be predicted more accurately.

[0150] The updating submodule is used to record the status at the next time point into the corresponding treatment plan of the corresponding patient to update the KOA database.

[0151] Furthermore, the KOA rehabilitation treatment decision module includes: a four-tuple data submodule and a dynamic recommendation submodule.

[0152] The four-tuple data submodule is used to obtain the historical status, treatment plan, cumulative reward value and status at the next time point of each patient from the updated KOA database to form four-tuple data;

[0153] The dynamic recommendation submodule is used to predict the treatment plan at a preset time point based on the four-tuple data, and recommend the treatment plan that the patient should choose within the preset time through a preset decision rule.

[0154] Second aspect

[0155] See also Figure 2 The present invention also provides an intelligent diagnosis and treatment method for knee osteoarthritis based on machine learning technology, comprising:

[0156] S001. Obtain the KOA database based on the multidimensional scale data of KOA patients;

[0157] S002, obtaining multiple feature quantities from the KOA database;

[0158] S003, extracting a key feature set from multiple feature quantities by adopting an iterative feature elimination strategy;

[0159] S004. Simulate the state of the patient at the next time point after the initial treatment plan is adopted according to the key feature set; record the state at the next time point under the corresponding treatment plan of the corresponding patient to update the KOA database;

[0160] S005. Based on the updated KOA database, recommend the treatment plan that the patient should choose within the preset time through the preset decision rules;

[0161] S006. Optimize the treatment plan by learning the optimal strategy function to obtain the optimal treatment plan.

[0162] See also Figure 3 , the extracting key feature set from multiple feature quantities by adopting an iterative feature elimination strategy includes: constructing and training a random forest model, and based on the random forest model, adopting an iterative feature elimination strategy to screen multiple feature quantities in the KOA database to obtain a key feature set;

[0163] The random forest model is constructed and trained, including:

[0164] S301, constructing an initial random forest model;

[0165] S302, using self-service sampling, select n samples from multiple feature quantities in the KOA database to form a training set; randomly select m features from the training set;

[0166] S303, using out-of-bag error estimation to calculate the importance score of each feature in the first training set;

[0167] The importance score calculation formula is as follows:

[0168]

[0169] Where:

[0170] : The importance score of each feature Xi;

[0171] : the number of samples;

[0172] : The number of samples correctly classified without feature Xi;

[0173] : The number of samples correctly classified after the feature Xi;

[0174] S304, sorting all features in descending order according to importance scores;

[0175] S305, using step-by-step back-off search to iteratively remove the feature with the lowest importance score in the first training set;

[0176] S306, calculating the classification accuracy of the random forest model after removing a feature each time;

[0177] S307. When the classification accuracy is stabilized at a preset value, stop removing features to obtain a key feature set, which is a set of remaining features after removing some features in the first training set according to S305 to S307.

[0178] Furthermore, the optimal treatment plan is obtained by optimizing the treatment plan through learning the optimal strategy function, which specifically includes:

[0179] S601, constructing a buffer; constructing a buffer to efficiently store and sample the KOA database; storing the patient's state transition data in the KOA database, i.e., four-tuple data (each patient's historical state, treatment plan, accumulated reward value, and state at the next time point) into the buffer;

[0180] S602, perform data enhancement based on the buffer; randomly sample a historical state from the buffer as a state sample, randomly match the historical state with the treatment plan, then simulate and predict the state of the patient at the next time point after the randomly matched treatment plan, and obtain a prediction reward based on the state at the next time point;

[0181] If the predicted reward is higher than the preset reward value, the historical state and treatment plan that match at this time are bound as enhanced new samples and added to the buffer;

[0182] S603, repeat S602 until the number of newly added samples reaches a preset threshold;

[0183] S604, updating the parameters of the strategy function in the preset decision rule according to the samples in the buffer, and recording the immediate feedback value of the reward function as a training curve;

[0184] The expression of the strategy function is as follows:

[0185]

[0186] Where:

[0187] : The historical status of the patient at time point t;

[0188] : Treatment options;

[0189] S605. Obtain the actual state of the patient in the sample of the buffer after the treatment plan is carried out during the observation period (from the next time point to the time point t), and evaluate the actual performance of the strategy according to the strategy function; by using the difference between the predicted state at the next time point and the actual state as the evaluation standard, and using the minimization objective function to observe the difference, the expression of minimizing the objective function is:

[0190]

[0191] Where:

[0192] : The treatment regimen selected at time point t;

[0193] T: the total number of time steps from the next time point to time t;

[0194] S606: Calculate and adopt a treatment plan based on the reward function The immediate feedback value of the reward received ;

[0195] S607, according to the instant feedback value , Patient status and treatment options Combining to form counterfactual decision trajectories , where T represents the total number of time steps;

[0196] The cumulative reward value The expression is:

[0197] ;

[0198] At the same time, according to the actual state, the actual cumulative reward value R is calculated through the reward function. The expression of the actual cumulative reward value R is:

[0199] ;

[0200] S608, by accumulating reward value Calculate the success rate with the actual cumulative reward value R;

[0201]

[0202] Where: = -R;

[0203] The highest success rate will be selected (accumulated reward value The treatment plan combination corresponding to the time when the difference between the actual cumulative reward value R is the smallest is the most optimal treatment plan.

[0204] This approach allows the potential effects of new strategies to be evaluated without actually changing the patient's treatment regimen, providing a scientific basis for further strategy optimization.

[0205] The above embodiments only express preferred implementation modes, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention.

Claims

1. Intelligent diagnosis and treatment system for knee osteoarthritis based on machine learning technology, characterized by: include: A data preprocessing module is used to obtain a KOA database based on multi-dimensional scale data of KOA patients; A feature extraction module is used to obtain multiple feature quantities in the KOA database; KOA feature screening module, used to extract key feature sets from multiple feature quantities using an iterative feature elimination strategy; The KOA rehabilitation treatment effect simulation module is used to simulate the patient's state at the next time point after adopting the initial treatment plan based on the key feature set; Record the status at the next time point under the corresponding treatment plan for the corresponding patient to update the KOA database; The KOA rehabilitation treatment decision module is used to recommend the treatment plan that the patient should choose within the preset time according to the updated KOA database through the preset decision rules; The intelligent diagnosis and treatment system for knee osteoarthritis further includes: a KOA intelligent treatment auxiliary decision module, which is used to optimize the KOA rehabilitation treatment decision module by learning the optimal strategy function to obtain the optimal treatment plan; The KOA intelligent treatment auxiliary decision module includes a reward submodule, a deep reinforcement submodule and a decision evaluation submodule; The reward submodule is used to evaluate the effectiveness of the treatment plan through a reward function to obtain an immediate feedback value; The deep reinforcement submodule is used to learn the optimal strategy function according to the immediate feedback value to obtain the cumulative reward value of each treatment plan; The decision evaluation submodule is used to output the optimal treatment plan according to the accumulated reward value; the optimal treatment plan is the treatment plan combination with the highest cumulative reward value combination score after the treatment plans are combined; The expression of the reward function is as follows: Where: : Instant feedback value; : KOA pain score at week t; : KOA pain score at week t+1; : The first threshold of KOA pain score; : The second threshold of KOA pain score; is a constant; The reward function includes three cases: when < To improve the situation, the reward function is calculated as: ; when = When , it is a stable situation, then compare Thresholds with KOA pain scores; like > ,but ; like ≦ ,but ; like < ≦ ,but ; when > When , it is a deteriorating situation, then the calculation formula of the reward function is; 。 2. The intelligent diagnosis and treatment system for knee osteoarthritis based on machine learning technology according to claim 1 is characterized by: The data preprocessing module includes: a data cleaning submodule, a missing value processing submodule, a category imbalance processing submodule and a normalization submodule; The data cleaning submodule is used to remove abnormal values ​​in the scale data to obtain normal value scale data; The missing value processing submodule is used to fill in the independent missing values ​​in the normal value scale data to obtain the whole scale data; The class imbalance processing submodule is used to obtain balanced scale data by resampling and cost sensitivity analysis of the whole scale data; The normalization submodule is used to process the balance scale data by data normalization to obtain the KOA database.

3. The intelligent diagnosis and treatment system for knee osteoarthritis based on machine learning technology according to claim 1 is characterized by: The KOA feature screening module includes: an initial random forest submodule and a random forest submodule; The initial random forest submodule is used to obtain a random forest submodule through KOA database training; The random forest submodule is used to adopt an iterative feature elimination strategy to screen multiple feature quantities in the KOA database to obtain a key feature set.

4. The intelligent diagnosis and treatment system for knee osteoarthritis based on machine learning technology according to claim 1 is characterized by: The KOA rehabilitation treatment effect simulation module includes: a sequence submodule, a history information encoding submodule, a multi-layer perception submodule and an update submodule; The sequence submodule is used to obtain time series disease data in the key feature set; the time series disease data includes the patient's historical status and treatment plan; The historical information encoding submodule is used to encode the historical status and treatment plan to obtain the historical information code; The multi-layer perception submodule is used to predict the patient's status at the next time point based on the historical information encoding; The updating submodule is used to record the status at the next time point into the corresponding treatment plan of the corresponding patient to update the KOA database.

5. The intelligent diagnosis and treatment system for knee osteoarthritis based on machine learning technology according to claim 1 is characterized by: The KOA rehabilitation treatment decision module includes: a four-tuple data submodule and a dynamic recommendation submodule; The four-tuple data submodule is used to obtain the historical status, treatment plan, cumulative reward value and status at the next time point of each patient from the updated KOA database to form four-tuple data; The dynamic recommendation submodule is used to predict the treatment plan for a preset time based on the four-tuple data, and recommend the treatment plan that the patient should choose within the preset time through a preset decision rule.

6. An intelligent diagnosis and treatment method for knee osteoarthritis based on machine learning technology, characterized in that: include: S001. Obtain the KOA database based on the multidimensional scale data of KOA patients; S002, obtaining multiple feature quantities from the KOA database; S003, extracting a key feature set from multiple feature quantities by adopting an iterative feature elimination strategy; S004. Simulate the state of the patient at the next time point after the initial treatment plan is adopted according to the key feature set; Record the status at the next time point under the corresponding treatment plan for the corresponding patient to update the KOA database; S005. Based on the updated KOA database, recommend the treatment plan that the patient should choose within the preset time through the preset decision rules; S006. Optimizing the treatment plan by learning the optimal strategy function to obtain the optimal treatment plan; The intelligent diagnosis and treatment method for knee osteoarthritis further includes: a KOA intelligent treatment auxiliary decision-making process for optimizing the treatment plan by learning the optimal strategy function to obtain the optimal treatment plan; The KOA intelligent treatment auxiliary decision-making process includes a reward sub-process, a deep reinforcement sub-process and a decision-making evaluation sub-process; The reward sub-process is used to evaluate the effectiveness of the treatment plan through a reward function to obtain an immediate feedback value; The deep reinforcement sub-process is used to learn the optimal strategy function according to the immediate feedback value to obtain the cumulative reward value of each treatment plan; The decision-making evaluation sub-process is used to output the optimal treatment plan according to the accumulated reward value; the optimal treatment plan is the treatment plan combination with the highest cumulative reward value combination score after the treatment plans are combined; The expression of the reward function is as follows: Where: : Instant feedback value; : KOA pain score at week t; : KOA pain score at week t+1; : The first threshold of KOA pain score; : The second threshold of KOA pain score; is a constant; The reward function includes three cases: when < To improve the situation, the reward function is calculated as: ; when = When , it is a stable situation, then compare Thresholds with KOA pain scores; like > ,but ; like ≦ ,but ; like < ≦ ,but ; when > When , it is a deteriorating situation, then the calculation formula of the reward function is; 。 7. The intelligent diagnosis and treatment method for knee osteoarthritis based on machine learning technology according to claim 6, characterized in that: The method of extracting a key feature set from multiple feature quantities by adopting an iterative feature elimination strategy includes: constructing and training a random forest model; based on the random forest model, adopting an iterative feature elimination strategy to screen multiple feature quantities in the KOA database to obtain a key feature set; The random forest model is constructed and trained, including: S301, constructing an initial random forest model; S302, using self-service sampling, randomly selecting m features from multiple features in the KOA database to form a first training set; S303, calculating the importance score of each feature in the first training set using out-of-bag error estimation; S304, sorting all features in descending order according to importance scores; S305, using step-by-step back-off search to iteratively remove the feature with the lowest importance score in the first training set; S306, calculating the classification accuracy of the random forest model after removing a feature each time; S307. When the classification accuracy is stabilized at a preset value, stop removing features to obtain a key feature set, which is a set of remaining features after removing some features in the first training set according to S305 to S307.