KOA rehabilitation treatment effect simulation method and system based on recurrent neural network

Abstract

The invention discloses a KOA rehabilitation treatment effect simulation method and system based on a recurrent neural network. A data preprocessing module is used for obtaining time sequence disease data; the time sequence disease data comprises a historical state of a patient and a plurality of treatment methods; the historical information coding module is used for coding the time sequence disease data to obtain historical information codes; the multi-layer sensing module is used for predicting the patient state of the patient at the next time point according to the historical information codes; the scheme decision module is used for determining a treatment method received by the patient according to the patient state at the next time point; the method has the advantages that simulation prediction of various treatment methods for the current illness state of the patient is completed, and the optimal treatment method is selected from the multiple treatment methods under the condition that real medical resources and medical time are not wasted.

Description

Technical Field

[0001] The present invention relates to the field of intelligent treatment technology, and in particular to a KOA rehabilitation treatment effect simulation method and system based on a recurrent neural network. Background Art

[0002] Auxiliary diagnosis system is a tool based on computer technology and artificial intelligence algorithms to help doctors diagnose and evaluate diseases. This system combines medical knowledge and technology, and uses technologies such as big data, machine learning, image processing, and natural language processing to analyze medical data and provide real-time advice and decision support to assist doctors in diagnosing diseases. Auxiliary diagnosis system can be used to solve complex problems in diagnosis, assist doctors in the decision-making process, and provide more comprehensive and accurate information about the patient's condition. It can automatically label, segment and interpret medical images, and assist doctors in analyzing and understanding abnormalities in images. In addition, auxiliary diagnosis system can also integrate and analyze patients' clinical data and provide information on disease probability, treatment recommendations, and prognosis evaluation. Auxiliary diagnosis system can reduce the workload of doctors, improve diagnostic accuracy and efficiency, reduce misdiagnosis rate, and provide better solutions and results for patients' treatment, while also contributing to the accumulation of medical research and knowledge.

[0003] However, existing technologies lack in-depth analysis of long-term, continuous time series data, resulting in the inability to accurately track the evolution of the disease. The system only provides decision-making advice to doctors at a single point in time, without taking into account the long-term development and changes of the disease. In addition, there are significant differences in the condition of patients with knee osteoarthritis and their response to treatment methods. Accurately mapping the dynamics of an individual's condition for personalized treatment costs doctors a lot of time and energy. If the development of the disease after illness can be predicted, such as the type of disease after a period of time and death after a few years, it will be very helpful for doctors to take what treatment methods in advance.

[0004] In view of the deficiencies in the prior art, the present invention proposes a KOA rehabilitation treatment effect simulation method and system based on a recurrent neural network, which utilizes the neural network to obtain more accurate prognosis effect simulation results. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a KOA rehabilitation treatment effect simulation method and system based on a recurrent neural network based on time series data, which overcomes the defect that the existing disease development trajectory prediction method cannot take into account both performance and interpretability. By constructing a state transfer function and a simulation function, it is accurately predicted that the patient's state at the next time point under a certain treatment method; thereby completing the simulation prediction of multiple treatment methods for the patient's current condition, and selecting the best treatment method from a large number of treatment methods without wasting real medical resources and medical time.

[0006] The purpose of the present invention is achieved through the following technical solutions: A KOA rehabilitation treatment effect simulation system based on a recurrent neural network, comprising:

[0007] A data preprocessing module, used to obtain time series disease data; the time series disease data includes a patient's historical status and multiple treatment methods;

[0008] A historical information encoding module, used for encoding time series disease data to obtain historical information encoding;

[0009] A multi-layer perception module predicts the patient's status at the next time point based on historical information encoding;

[0010] The treatment plan decision module determines the treatment method the patient will receive based on the patient's status at the next time point.

[0011] The system of the present invention simulates and predicts the health state of the patient at the next time point when the patient receives the corresponding treatment plan by inputting the patient's current health state and the different treatment methods he / she has received.

[0012] Furthermore, the historical information encoding module in the historical information encoding module is t The expression is as follows:

[0013] c t =GRU(s t-ω:t ,a t-ω:t )

[0014] Where:

[0015] ω: historical time window;

[0016] t: start time point;

[0017] s t : The historical status of the patient at time point t within the historical time window;

[0018] a t : Treatment methods;

[0019] s t-ω:t : The sequence of patient states from time point t-ω to t;

[0020] a t-ω:t : The sequence of treatment methods from time point t-ω to t.

[0021] By encoding historical information, it is ensured that the information that requires long-term dependence can be effectively transmitted.

[0022] Furthermore, the system also includes a standardization module, which is used to perform standardization processing on two variables, the historical state and the treatment method, in the historical information coding before the historical information coding generated by the historical information coding module is input into the multi-layer perception module; the formula for the standardization processing transformation is:

[0023]

[0024] Where:

[0025] Z(x): standardized variable;

[0026] x: variable;

[0027] σ: standard deviation of the state variable

[0028] μ: mean of the state variable.

[0029] The purpose of standardizing the two variables of historical status and treatment method is to improve the generalization ability and prediction accuracy of the system.

[0030] Furthermore, the loss function of the multi-layer perception module is expressed as:

[0031]

[0032] Where:

[0033] y itd : The true value of the d-th state variable of the ith patient at time point t,

[0034] The predicted value of the dth state variable for the ith patient at time point t;

[0035] N: total number of patients;

[0036] Ti: total number of time points for the ith patient;

[0037] D: total number of state variables;

[0038] By using the loss function to reduce the difference between the predicted value, that is, the patient's state at the next time point, and the actual value, the patient's state at the next time point can be predicted more accurately.

[0039] Further, the system also includes: using a sensitivity analysis module for setting the historical time window ω to 0, 1, 2, and 4, respectively, to train the KOA rehabilitation treatment effect simulation system;

[0040] When the historical time window ω=0, the historical information encoding module is replaced by a multi-layer perception module.

[0041] Furthermore, the system further comprises:

[0042] The success rate evaluation module is used to obtain the prediction success rate by calculating the mean absolute error and the mean square error, and use the prediction success rate as an indicator of system performance;

[0043] The formula for the mean absolute error is:

[0044]

[0045] The formula for the mean square error is:

[0046]

[0047] Where:

[0048] N: total number of test samples;

[0049] MAE: mean absolute error;

[0050] MSE: mean square error;

[0051] Represents the predicted value of the i-th sample;

[0052] y i : represents the true value of the i-th sample;

[0053] The formula for the success rate is:

[0054]

[0055] Where:

[0056] The jth predicted value of the i-th sample;

[0057] y ij : The jth true value of the ith sample;

[0058] max j : The maximum value of the jth index;

[0059] min j : The minimum value of the j-th indicator.

[0060] Furthermore, the prediction success rate is the difference between the predicted value and the true value, and the extreme value range of this indicator is 0 to 10%.

[0061] The present invention also provides a method for simulating the effect of KOA rehabilitation treatment based on a recurrent neural network, comprising:

[0062] S1. Obtain time series disease data; the time series disease data includes the patient's historical status s t and multiple treatmentsa t ;

[0063] S2, based on the patient's historical status s at time point t t and treatment methodsa t Get historical information code c t ;

[0064] S3, encoding c according to historical information t Construct a simulation function f(s t ,a t ,c t ) to predict the patient's status at the next time point;

[0065] S4, obtaining a patient status sequence and a treatment method sequence from time point t-ω to time point t by predicting the patient's status at the next time point; ω is a historical time window;

[0066] S5. Determine the treatment method to be received by the patient from the patient's state sequence and treatment method sequence according to the patient's state at the next time point.

[0067] The present invention has the following advantages:

[0068] (1) The present invention obtains time series disease data; uses a historical information encoding module to encode the time series disease data to obtain historical information encoding, and predicts the patient's status at the next time point by encoding the historical information; maps continuous variables and discrete variables to the same new variable space, and introduces the correlation between neural network connectivity evaluation features, thereby obtaining the causal relationship between the patient's status and the treatment method; thereby obtaining a more accurate simulation prediction result.

[0069] (2) The present invention constructs state transfer functions and simulation functions to accurately predict the patient's state at the next time point under the action of a certain treatment method; thereby completing the simulation prediction of multiple treatment methods for the patient's current condition, and selecting the best treatment method from a variety of treatment methods without wasting real medical resources and medical time; by simulating and predicting the results of different treatment methods and health status combinations, the possible results of these treatment methods are indirectly evaluated, which allows researchers to evaluate and optimize the selection of new treatment methods without directly applying them to patients; this not only protects patients from potential risks, but also accelerates the development process of new treatment strategies.

[0070] (3) The present invention constructs the prediction success rate through the average mean square error and the mean square error, and adjusts the parameters of the system according to the success rate, thereby obtaining an accurate prognosis effect simulation prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 is a system structure diagram of the present invention;

[0072] Figure 2 Schematic diagram of the method steps of the invention. DETAILED DESCRIPTION

[0073] 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.

[0074] 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.

[0075] 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.

[0076] First aspect

[0077] See also Figure 1 , KOA rehabilitation treatment effect simulation system based on recurrent neural network, including:

[0078] A data preprocessing module, used to obtain time series disease data; the time series disease data includes a patient's historical status and multiple treatment methods;

[0079] A historical information encoding module, used for encoding time series disease data to obtain historical information encoding;

[0080] A multi-layer perception module predicts the patient's status at the next time point based on historical information encoding;

[0081] The treatment plan decision module determines the treatment method the patient will receive based on the patient's status at the next time point.

[0082] The system of the present invention simulates and predicts the health state of the patient at the next time point when the patient receives the corresponding treatment plan by inputting the patient's current health state and the different treatment methods he / she has received.

[0083] The KOA rehabilitation treatment effect simulation system using recurrent neural networks can simulate in detail the complex relationship between the patient's status and the treatment method, achieving several key goals: First, the KOA rehabilitation treatment effect simulation system using recurrent neural networks provides a prediction of the patient's status at the next time point, which has a key impact on the formulation of long-term treatment plans; the model can predict the changing trend of the patient's pain level and mobility in the next few weeks under a specific treatment plan, providing data support for medical experts to help them choose the best treatment method; secondly, the KOA rehabilitation treatment effect simulation system using recurrent neural networks helps develop personalized treatment plans by simulating the state transitions caused by treatment methods; considering that patients with knee osteoarthritis have significant differences in their condition and response to treatment, accurately mapping the individual's condition dynamics is an important tool for personalized treatment design.

[0084] Furthermore, the historical information encoding module is used to output the historical information encoding c according to the dynamic characteristics and historical dependence of the historical state and treatment method according to the time series. t , historical information encoding c t The expression is as follows:

[0085] c t =GRU(s t-ω:t ,a t-ω:t )

[0086] Where:

[0087] ω: historical time window;

[0088] t: start time point;

[0089] s t : The historical status of the patient at time point t within the historical time window;

[0090] a t : Treatment methods;

[0091] s t-ω:t : The sequence of patient states from time point t-ω to t;

[0092] a t-ω:t : The sequence of treatment methods from time point t-ω to t.

[0093] By encoding historical information, it is ensured that the information that requires long-term dependence can be effectively transmitted.

[0094] Furthermore, the system also includes a standardization module, which is used to perform standardization processing on two variables, the historical state and the treatment method, in the historical information coding before the historical information coding generated by the historical information coding module is input into the multi-layer perception module; the formula for the standardization processing transformation is:

[0095]

[0096] Where:

[0097] Z(x): standardized variable;

[0098] x: variable;

[0099] σ: standard deviation of the state variable

[0100] μ: mean of the state variable.

[0101] The purpose of standardizing the two variables of historical status and treatment method is to improve the generalization ability and prediction accuracy of the system.

[0102] Furthermore, the loss function of the multi-layer perception module is expressed as:

[0103]

[0104] Where:

[0105] y itd : The true value of the d-th state variable of the ith patient at time point t,

[0106] The predicted value of the dth state variable for the ith patient at time point t;

[0107] N: total number of patients;

[0108] Ti: total number of time points for the ith patient;

[0109] D: total number of state variables;

[0110] 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 loss function, the system can capture and understand the complex interdependencies and dynamic changes between state variables, and can more accurately predict the state at future time points.

[0111] Furthermore, the system also includes a sensitivity analysis module for setting the historical time window ω to 0, 1, 2 and 4 respectively to train the KOA rehabilitation treatment effect simulation system; when the historical time window ω=0, the historical information encoding module is replaced by a multi-layer perception module.

[0112] Furthermore, the system further comprises: a success rate evaluation module, which is used to obtain the prediction success rate by calculating the mean absolute error and the mean square error, and use the prediction success rate as an indicator of system performance;

[0113] The formula for the mean absolute error is:

[0114]

[0115] The formula for the mean square error is:

[0116]

[0117] Where:

[0118] N: total number of test samples;

[0119] MAE: mean absolute error;

[0120] MSE: mean square error;

[0121] Represents the predicted value of the i-th sample;

[0122] y i : represents the true value of the i-th sample;

[0123] The formula for the success rate is:

[0124]

[0125] Where:

[0126] The jth predicted value of the i-th sample;

[0127] y ij : The jth true value of the ith sample;

[0128] max j : The maximum value of the jth index;

[0129] min j : The minimum value of the j-th indicator.

[0130] Furthermore, the prediction success rate is the difference between the predicted value and the true value, and the extreme value range of the difference between the predicted value and the true value is 0-10%.

[0131] Second aspect

[0132] See also Figure 2 The present invention also provides a method for simulating the effect of KOA rehabilitation treatment based on a recurrent neural network, comprising:

[0133] S1. Obtain time series disease data; the time series disease data includes the patient's historical status s t and multiple treatmentsa t ;

[0134] According to the historical status t and multiple treatmentsa tConstruct the patient state transition process function P(s t+1 |s t ,a t ), obtain the patient's historical status s from time point t t The patient's status s at time t+1 t+1 , through the patient state transition process function to capture the temporal dependency of the patient's historical state, and obtain the association between the historical state and the treatment method;

[0135] S2. Obtain historical information code c based on patient time series disease data t ;

[0136] The GRU model is used to encode the patient's historical state s from time point t-ω to time point t t and treatment methodsa t The historical trajectory of the historical information is obtained by t , historical information encoding c t The expression is as follows:

[0137] c t =GRU(s t-ω:t ,a t-ω:t )

[0138] Where:

[0139] ω: historical time window;

[0140] t: start time point;

[0141] s t : The historical status of the patient at time point t within the historical time window;

[0142] a t : Treatment methods;

[0143] s t-ω:t : The sequence of patient states from time point t-ω to t;

[0144] a t-ω:t : The sequence of treatment methods from time point t-ω to t;

[0145] S3, encoding c according to historical information t Construct a simulation function f(s t ,a t ,c t ) to predict the patient's status at the next time point;

[0146] Get the historical information code c t After that, the simulation function f(s) is constructed through the multi-layer perception module. t ,a t ,c t) to predict the patient’s status at the next time point and encode the historical information c t As input, the output predicts the patient's state s at time point t+1 t+1 ;

[0147] The loss function of the multi-layer perception module is expressed as:

[0148]

[0149] Where:

[0150] y itd : The true value of the d-th state variable of the ith patient at time point t,

[0151] The predicted value of the dth state variable for the ith patient at time point t;

[0152] N: total number of patients;

[0153] Ti: total number of time points for the ith patient;

[0154] D: total number of state variables;

[0155] Through the loss function, the predicted value is reduced, that is, the patient's state s at the next time point, that is, at time point t+1 t+1 The difference from the actual value allows the system to capture and understand the complex interdependencies and dynamic changes between state variables, and can more accurately predict the state of the exchange at a future point in time;

[0156] S4, obtaining a patient status sequence and a treatment method sequence from time point t-ω to time point t by predicting the patient's status at the next time point; ω is a historical time window;

[0157] S5. Determine the treatment method to be received by the patient from the patient's state sequence and treatment method sequence according to the patient's state at the next time point.

[0158] 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. KOA rehabilitation treatment effect simulation system based on recurrent neural network, characterized by: include: A data preprocessing module, used to obtain time series disease data; the time series disease data includes a patient's historical status and multiple treatment methods; A historical information encoding module, used for encoding time series disease data to obtain historical information encoding; A multi-layer perception module predicts the patient's status at the next time point based on historical information encoding; The treatment plan decision module selects and determines the treatment method for the patient based on the patient's status at the next time point.

2. The method and system for simulating the effect of KOA rehabilitation treatment based on a recurrent neural network according to claim 1, characterized in that: The historical information encoding module in the historical information encoding module is t The expression is as follows: w t =GRU(s t-ω:t ,the t-ω:t ) Where: ω: historical time window; t: time point; s t : The historical status of the patient at time point t within the historical time window; a t : Treatment methods; s t-ω:t : The sequence of patient states from time point t-ω to t; a t-ω:t : The sequence of treatment methods from time point t-ω to t.

3. The KOA rehabilitation treatment effect simulation system based on recurrent neural network according to claim 2 is characterized in that: The system further comprises: The standardization module is used to perform standardization on the two variables of historical status and treatment method in the historical information coding before inputting the historical information coding generated by the historical information coding module into the multi-layer perception module; the formula for the standardization transformation is: Where: Z(x): standardized variable; x: variable; σ: standard deviation of the variable; μ: mean of the variable.

4. The KOA rehabilitation treatment effect simulation system based on recurrent neural network according to claim 2 is characterized in that: The loss function of the multi-layer perception module is expressed as: Where: y itd : The true value of the d-th state variable of the ith patient at time point t, The predicted value of the dth state variable for the ith patient at time point t; N: total number of patients; Ti: total number of time points for the ith patient; D: The total number of state variables.

5. The KOA rehabilitation treatment effect simulation system based on recurrent neural network according to claim 2 is characterized in that: The system also includes: A sensitivity analysis module was used to set the historical time window ω to 0, 1, 2, and 4, respectively, to train the KOA rehabilitation treatment effect simulation system.

6. The KOA rehabilitation treatment effect simulation system based on recurrent neural network according to claim 1 is characterized in that: The system further comprises: The success rate evaluation module is used to obtain the prediction success rate by calculating the mean absolute error and the mean square error, and use the prediction success rate as an indicator of system performance; The formula for the mean absolute error is: The formula for the mean square error is: Where: N: total number of patients; MAE: mean absolute error; MSE: mean square error; represents the predicted value of the i-th patient; y i : represents the true value of the i-th patient; The formula for the success rate is: Where: The jth predicted value for the i-th patient; y ij : The jth true value of the ith patient; max j : The maximum value of the jth index; min j : The minimum value of the j-th indicator.

7. The KOA rehabilitation treatment effect simulation system based on recurrent neural network according to claim 6 is characterized in that: The success rate of the prediction is the difference between the predicted value and the true value, and the extreme value range of the difference between the predicted value and the true value is 0-10%.

8. The KOA rehabilitation treatment effect simulation method based on recurrent neural network is characterized by: include: S1. Obtain time series disease data; the time series disease data includes the patient's historical status s t and multiple treatmentsa t ; S2, based on the patient's historical status s at time point t t and treatment methodsa t Get historical information code c t ; S3, encoding c according to historical information t Construct a simulation function f(s t ,a t ,c t ) to predict the patient's status at the next time point; S4, obtaining a patient status sequence and a treatment method sequence from time point t-ω to time point t by predicting the patient's status at the next time point; ω is a historical time window; S5. Determine the treatment method to be received by the patient from the patient's state sequence and treatment method sequence according to the patient's state at the next time point.

Images