Medical decision interpretation method and system

Abstract

The present invention relates to a medical decision interpretation method and system. The method comprises: collecting patient data through a database and a questionnaire; processing the patient data, inputting the processed patient data into a classification model to extract data features, and making medical prediction recommendations based on the data features; using a SHAP model and a white box model to interpret the classification models respectively, and combining the interpretation results for visual comparison, and analyzing the combined data features to obtain the final decision interpretation results. The SHAP model and the white box model are combined to interpret the classification model, providing sufficient support for the medical prediction recommendations output by the classification model, overcoming the unexplainable problems in the traditional medical recommendation system, providing patients with sufficient explainable information, making it easier for patients to accept the recommendation results, so as to achieve the purpose of predicting and preventing diseases, and improving the user's trust in the system recommendation results, which is of great significance to medical recommendations.

Description

Technical Field

[0001] The present invention relates to the field of medical health information technology, and in particular to a medical decision interpretation method and system. Background Art

[0002] With the development of society, people's living standards are constantly improving, and more and more people pay attention to their physical health. In recent years, the proportion of the elderly population in my country has continued to increase rapidly, becoming the main demand group for medical services, which has aroused widespread concern in society. Chronic diseases of the elderly are not only difficult to cure, but also take a long time to treat, and are prone to relapse, requiring a variety of treatment methods. However, there is a shortage of experienced doctors in this field, which brings great challenges to medical diagnosis. The medical decision system can make recommendations based on the patient's medical data and current diagnostic reports, which can reduce the difficulty of treating chronic diseases in the elderly and enable more elderly patients to receive medical help in time. Although the emergence of medical recommendation systems is helpful in treating chronic diseases, it cannot replace the role of doctors. The system can provide special advice to patients and doctors, reminding patients to correct bad behaviors during the prevention and treatment stages of the disease. At the same time, the use of medical decision systems can also reduce the treatment time and resource investment required in emergency situations.

[0003] The introduction of medical decision-making systems effectively solves the problem of data overload caused by information explosion, allowing users to quickly obtain the required information from huge amounts of data, thereby saving time. At present, many methods, including collaborative filtering technology, naive Bayes, decision trees, random forests, multi-layer perceptrons, and other technologies have been applied to medical decision-making systems. Although these methods can help users find the items they may need, ideally, medical decision-making systems should be understandable and explainable, and be able to explain the logic of the decision in detail to all relevant parties in order to win the trust of doctors and patients. Traditional medical decision-making systems do not explain the reasons for making recommendations, so patients have a low degree of acceptance of the recommendations made by medical decision-making systems. Summary of the invention

[0004] Based on this, in order to solve the above technical problems, a medical decision interpretation method and system are provided, which can explain the recommendations made and improve patient acceptance.

[0005] A medical decision interpretation method, comprising:

[0006] Collect patient data through database and questionnaire;

[0007] Performing data cleaning and data standardization on the patient data to obtain processed patient data;

[0008] Inputting the processed patient data into a classification model, extracting data features through the classification model, and making medical prediction recommendations based on the data features; wherein logistic regression, support vector machine, random forest, gradient boosting tree, artificial neural network, k-nearest neighbor, and Bayesian network are selected as classification models;

[0009] The classification models are explained respectively using the SHAP model and the white box model, and the explanation results are combined for visual comparison, and the data features are combined for analysis to obtain the final decision explanation results.

[0010] In one embodiment, the method further comprises:

[0011] Randomly dividing the patient data into a training set, a validation set, and a test set, and performing data cleaning and data standardization on the divided patient data to obtain processed training set, validation set, and test set data;

[0012] The training process of the classification model includes:

[0013] The training set data is mapped to an initial classification model, the initial classification model is trained by a Nadam algorithm, the training set data is changed, the model parameters of the initial classification model are adjusted, and a classification model is obtained; and the classification model is evaluated by using a k-fold cross validation technique.

[0014] In one embodiment, the training of the initial classification model by using the Nadam algorithm and adjusting the model parameters of the initial classification model include:

[0015] Calculating the gradient of the current parameter point in the initial classification model;

[0016] Calculate the first-order moment estimate and the second-order moment estimate of the gradient using exponential moving average;

[0017] Correcting the first-order moment estimation and the second-order moment estimation respectively to eliminate deviations;

[0018] The model parameters of the initial classification model are adjusted using the corrected first-order moment estimate and second-order moment estimate.

[0019] In one embodiment, the evaluating the classification model using a k-fold cross validation technique comprises:

[0020] Acquire an evaluation data set, and randomly divide the evaluation data set into a plurality of data subsets;

[0021] Select one from the plurality of data subsets as a test subset and the rest as training subsets;

[0022] Training and evaluating the classification model using the test subset and the training subset, and recording the accuracy, precision, and recall of the training and evaluation;

[0023] The average value of the performance index of the classification model is calculated according to the accuracy, precision and recall rate, and the average value of the performance index is used as the evaluation result of the classification model.

[0024] In one embodiment, the data cleaning process of the patient data includes:

[0025] The patient data are processed for missing values, detected and processed for outliers, and processed for duplicates to obtain processed preliminary data;

[0026] The text data in the processed preliminary data is cleaned by using text vector natural language processing technology to obtain processed patient data.

[0027] In one embodiment, the step of performing data standardization on the patient data includes:

[0028] The batch normalization technology in deep learning is used to perform data normalization on the patient data.

[0029] In one embodiment, the classification model is explained using a SHAP model, including:

[0030] Inputting the processed patient data into a classification model, extracting data features through the classification model, and assigning a SHAP value to each feature of each data point using a SHAP model;

[0031] Sorting each feature according to the SHAP value to implement interpretation analysis of the classification model;

[0032] The SHAP value is used to represent the contribution value of the feature.

[0033] In one of the embodiments, the white box model includes decision trees, logistic regression, and decision rules.

[0034] A medical decision interpretation system, comprising:

[0035] Data collection module, used to collect patient data through database and questionnaire;

[0036] A data processing module, used for performing data cleaning and data standardization on the patient data to obtain processed patient data;

[0037] A medical decision-making module, used for inputting the processed patient data into a classification model, extracting data features through the classification model, and making medical prediction recommendations based on the data features; wherein logistic regression, support vector machine, random forest, gradient boosting tree, artificial neural network, k-nearest neighbor, and Bayesian network are selected as classification models;

[0038] The decision explanation module is used to use the SHAP model and the white box model to explain the classification model respectively, and combine the explanation results for visual comparison, and analyze the data features to obtain the final decision explanation result.

[0039] The above-mentioned medical decision explanation method and system use a combination of the SHAP model and the white-box model to explain the classification model, providing sufficient support for the medical prediction recommendations output by the classification model, overcoming the unexplainable problems in the traditional medical recommendation system, and providing patients with sufficient explainable information, making it easier for patients to accept the recommendation results, so as to achieve the purpose of predicting and preventing diseases, and improving users' trust in the system recommendation results, which is of great significance to medical recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A diagram of an application environment of a medical decision explanation method in one embodiment;

[0041] Figure 2 is a flowchart of a medical decision interpretation method in one embodiment;

[0042] Figure 3 A schematic diagram of model training time for a heart disease dataset in one embodiment;

[0043] Figure 4 A schematic diagram of model training time for a diabetes dataset in one embodiment;

[0044] Figure 5 Schematic diagram of kernel interpretation for the heart disease dataset;

[0045] Figure 6 Schematic diagram of kernel interpretation for the diabetes dataset;

[0046] Figure 7 Schematic diagram for explaining the features of the decision tree for heart disease;

[0047] Figure 8 Diagram showing the interpretation of the features of the diabetes decision tree

[0048] Fig. 9 A structural block diagram of a medical decision explanation system in one embodiment;

[0049] Fig.10It is a structural block diagram of a medical decision explanation system in another embodiment;

[0050] Fig.11 It is a schematic diagram of the operation of a medical decision explanation system in one embodiment;

[0051] Fig.12 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] The medical decision interpretation method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Figure 1 As shown, the application environment includes a computer device 110. The computer device 110 can collect patient data through a database and a questionnaire; the computer device 110 can perform data cleaning and data standardization on the patient data to obtain processed patient data; the computer device 110 can input the processed patient data into a classification model, extract data features through the classification model, and make medical prediction recommendations based on the data features; wherein logistic regression, support vector machine, random forest, gradient boosting tree, artificial neural network, k-nearest neighbor, and Bayesian network are selected as classification models; the computer device 110 can use the SHAP model and the white box model to interpret the classification model respectively, and combine the interpretation results for visual comparison, and analyze the data features to obtain the final decision interpretation result. Among them, the computer device 110 can be, but is not limited to, various personal computers, laptops, smart phones, robots, unmanned aerial vehicles, tablet computers and other devices.

[0054] In one embodiment, Figure 2 As shown, a medical decision interpretation method is provided, comprising the following steps:

[0055] Step 202: Collect patient data through a database and a questionnaire.

[0056] The computer device can collect patient data from a database and in combination with a questionnaire. The database can be a hospital database, which can contain medical records. The medical records can be obtained from a medical institution and can include clinical diagnosis, test results, treatment plans, various indicators and other data. The questionnaire can be pre-designed to collect patient data directly from individuals, and the questionnaire can cover aspects such as personal lifestyle, diet, exercise habits, family medical history, etc.

[0057] Specifically, patient data may include basic information, clinical diagnosis and various indicators, and other information collected through questionnaires. Basic information includes age, gender, height, and weight; clinical diagnosis and various indicators include cholesterol level, whether cholesterol tests have been performed, whether stroke has occurred, blood sugar level, systolic blood pressure, and diastolic blood pressure; other information collected through questionnaires includes smoking, drinking, daily fruit consumption, and exercise.

[0058] Step 204, performing data cleaning and data standardization processing on the patient data to obtain processed patient data.

[0059] After acquiring the patient data, the computer device can process the patient data, which may specifically include data cleaning processing and data standardization processing, so as to obtain processed patient data.

[0060] Step 206, input the processed patient data into the classification model, extract the data features through the classification model, and make medical prediction recommendations based on the data features; among them, logistic regression, support vector machine, random forest, gradient boosting tree, artificial neural network, k-nearest neighbor, and Bayesian network are selected as classification models.

[0061] Step 208, using the SHAP model and the white box model to interpret the classification model respectively, and combining the interpretation results for visual comparison, and analyzing the combined data features to obtain the final decision interpretation result.

[0062] Among them, SHAP is a method to explain the output of machine learning models. It uses game theory and local explanations. It belongs to the framework of post-explanation, has better consistency, and can be used to explain various black box models. Its core concept is the Shapley value, which is a method to describe the "weight" or "importance" of a specific feature when the model predicts a specific data point; the Shapley value can be used to quantify the local explanation of the model and answer why such a decision was made for this sample. Specifically, it calculates the marginal contribution of each feature, that is, the difference between the contribution when this feature is added and the contribution before this feature is added, which is the marginal contribution of this feature.

[0063] By comparing the main characteristic factors obtained by the SHAP model and the white-box model and jointly analyzing the characteristics, the comprehensive contribution characteristics are obtained, which can reflect the importance of the characteristics in the medical prediction recommendation results.

[0064] In this embodiment, the SHAP model is combined with the white box model to explain the classification model, providing sufficient support for the medical prediction recommendations output by the classification model, overcoming the unexplainable problems in the traditional medical recommendation system, providing patients with sufficient explainable information, making it easier for patients to accept the recommendation results, so as to achieve the purpose of predicting and preventing diseases, and improving users' trust in the system recommendation results, which is of great significance to medical recommendations.

[0065] In one embodiment, a medical decision interpretation method provided may also include a process of training a classification model, the specific process including: randomly dividing patient data into a training set, a validation set, and a test set, and performing data cleaning and data standardization on the divided patient data to obtain processed training set, validation set, and test set data; mapping the training set data to an initial classification model, training the initial classification model using a Nadam algorithm, changing the training set data, adjusting the model parameters of the initial classification model, and obtaining a classification model; and evaluating the classification model using a k-fold cross-validation technique.

[0066] In this embodiment, the processed training set data is mapped to the initial classification model. By changing the data in the training set, the model parameters are continuously adjusted so that the model can better fit the training parameters; the model is trained using the Nadam algorithm to improve the adaptability of the model.

[0067] For example, by repeatedly changing the training set data and the test set data and constantly adjusting the parameters of the model, the test accuracy of the seven classification models on the heart disease data set and the diabetes data set were obtained respectively: for the training and test performance accuracy of the heart disease data set, the accuracy of the test set and the gradient boosting algorithm corresponding to the support vector machine was basically consistent with the accuracy of the training set, the accuracy of the test set was relatively high, and the training performance was better; the accuracy of the artificial neural network, Bayesian network and k-nearest neighbor algorithm was relatively low; for the training and test accuracy of the diabetes data set, the accuracy of the test set of the support vector machine, random forest and logistic regression was relatively high.

[0068] After training with the Nadam algorithm, the training speed and effect of the neural network were improved by improving the calculation method of momentum and introducing an adaptive learning rate, which slightly improved the test accuracy of each classification model. Among them, the test accuracy of the artificial neural network was the most obvious.

[0069] It can be further seen that when the number of samples in the data set is very small, the error rates of the training set data and the test set data of the seven classification models are high, and the training accuracy error is large. From the difference between the training accuracy and the test accuracy, it can also be seen that relatively complex models have higher stability.

[0070] like Figure 3 , Figure 4 As shown in the figure, the model training time of artificial neural network, k-nearest neighbor algorithm and Bayesian network for heart disease dataset and diabetes dataset is longer than that of other classification algorithms. Figure 3 Model training time for the heart disease dataset, Figure 4 The model training time for the diabetes dataset. Therefore, it is inferred that the more complex the model, the longer the training time and the higher the corresponding training cost.

[0071] In one embodiment, a medical decision interpretation method provided may also include a process of training an initial classification model using a Nadam algorithm, the specific process including: calculating the gradient of the current parameter point in the initial classification model; using exponential moving average to calculate the first-order moment estimate and second-order moment estimate of the gradient; correcting the first-order moment estimate and second-order moment estimate respectively to eliminate bias; and using the corrected first-order moment estimate and second-order moment estimate to adjust the model parameters of the initial classification model.

[0072] Specifically, in this embodiment, the steps of using the algorithm to train the model can be divided into the steps of calculating the gradient, calculating the first-order moment estimate (Nesterov momentum term), calculating the second-order moment estimate (adaptive learning rate term), correcting the deviation, and Nesterov updating. Among them, calculating the gradient is mainly to calculate the gradient of the current parameter point; calculating the first-order moment estimate can also be called calculating the Nesterov momentum term, which is the first-order moment estimate of the gradient calculated using the exponential moving average, that is, the moving average of the gradient; calculating the second-order moment estimate can also be called the adaptive learning rate term, which is the second-order moment estimate of the gradient calculated using the exponential moving average, that is, the moving average of the square of the gradient; correcting the deviation is to correct the first-order and second-order moment estimates to eliminate the deviation during the initial training; Nesterov update is to use the corrected first-order and second-order moment estimates to update the model parameters. Specifically, the update rule can be expressed as: Among them, θ is the model parameter, is the learning rate, m is the first-order moment estimate, v is the second-order moment estimate, t is the time step, σ is a constant added for numerical stability, and its value is very small. β1 is the exponential decay rate of momentum and second-order moment estimate.

[0073] In one embodiment, a medical decision interpretation method provided may also include a process of evaluating a model using a k-fold cross-validation technique, the specific process including: obtaining an evaluation data set, and randomly dividing the evaluation data set into several data subsets; selecting one from the several data subsets as a test subset, and the rest as training subsets; training and evaluating the classification model through the test subset and the training subset, and recording the accuracy, precision, and recall rate of the training and evaluation; calculating the average performance index of the classification model based on the accuracy, precision, and recall rate, and using the average performance index as the evaluation result of the classification model.

[0074] In this embodiment, the evaluation data set can be randomly divided into three data subsets of similar size. The model can be trained and evaluated three times using the three data subsets, each time one of the data subsets is selected as the test subset and the remaining two are used as training subsets. By recording the accuracy, precision and recall rate of each training and evaluation, the average value of the performance indicators is calculated as an evaluation of the model performance.

[0075] That is, after determining the accuracy and training time of the model, the model effect is evaluated using accuracy, precision, recall rate and F1 score. In order to obtain more stable and reliable model performance evaluation indicators, the k-fold cross-validation method is used to evaluate the average value of k evaluation results to improve robustness. Under the premise of ensuring accuracy and precision, the recall rate is improved to obtain the final prediction result of the model.

[0076] In one embodiment, a medical decision interpretation method provided may also include a data cleaning process, the specific process including: performing missing value processing, outlier detection and processing, and duplicate item processing on patient data to obtain processed preliminary data; using text vector natural language processing technology to perform data cleaning on the text data in the processed preliminary data to obtain processed patient data.

[0077] Specifically, when processing missing values, the interpolation technology of the deep learning model can be used to more accurately estimate and process missing values; when detecting and processing outliers, the anomaly detection algorithm based on deep learning can be applied to more accurately identify and process outliers; when processing duplicate items, deep learning methods can be used to detect and process duplicate items to ensure that there is no duplicate or redundant data in the data set.

[0078] There is still text data in the processed preliminary data. For the text data, text vector natural language processing (NLP) technology can be used, including removing stop words, stem extraction and entity recognition, to better process and clean the text information.

[0079] In one embodiment, a medical decision interpretation method provided may also include a data standardization process, which specifically includes: using batch normalization technology in deep learning to perform data standardization on patient data. Batch normalization technology in deep learning is used to ensure that the input of the network layer remains standardized, thereby improving the stability and speed of model training.

[0080] In one embodiment, a medical decision interpretation method provided may also include a process of using a SHAP model to interpret a classification model, the specific process including: inputting processed patient data into the classification model, extracting data features through the classification model, and using the SHAP model to assign a SHAP value to each feature of each data point; sorting each feature according to the SHAP value to achieve an interpretation analysis of the classification model; wherein the SHAP value is used to represent the contribution value of the feature.

[0081] The SHAP model mainly attributes the output value to the contribution value of each feature, and measures the impact of each feature on the final output value by calculating the contribution value of each feature, which may be a positive contribution or a negative contribution. Specifically, after the processed patient data is input into the classification model, the SHAP model assigns a SHAP value to each feature of each data point, and the SHAP value indicates the contribution of the feature to the prediction result; finally, each feature is sorted by the SHAP value, and the model is analyzed in an interpretable manner to provide principle support. The sample size in SHAP is a factor that affects the expected value and average prediction of SHAP. The number of samples can be reasonably controlled to achieve a more accurate interpretation effect.

[0082] SHAP is used to achieve the interpretability of relatively complex models, obtain the importance ranking of each feature to the calculation model results, and give the contribution coefficient. In short, the contribution value of each feature is calculated to measure the impact of the feature on the final output value. In the SHAP model, not only can the exact reasons for the explanation results and the degree of certainty of the results be seen, but also by obtaining an overview of the entire data set, you can understand the contribution of each feature to the results of different categories of patients throughout the model process.

[0083] In this embodiment, each model is analyzed for interpretability through SHAP, and the classification model interpretation is experimented through the SHAP model. The experimental results are as follows: Figure 5 As shown, Figure 5 This is a diagram of the kernel explanation for the heart disease dataset. For the prediction explanation of heart disease, the high blood pressure feature contributes the most, followed by age and body mass index, and then health index and smokers. From the smokers, we can see that these features are the most important for heart disease, which can remind patients to pay attention to their habits in these aspects in the future to reduce the possibility of heart disease. The experimental results of the diabetes dataset are shown in Figure 2. Figure 6 As shown, Figure 6 Illustration of the kernel interpretation for the diabetes dataset, where glucose, body mass index, and age are the main contributing factors to diabetes.

[0084] In one embodiment, the calculation formula used by the SHAP model is: Among them, f(x) represents the predicted value of the sample in the decision tree, z i '∈{0,1} M Indicates how many of the M features are included in the decision path of the sample. For a sample, if feature k is not in its decision path, the SHAP value of the corresponding feature is 0, that is, φ k = 0, indicating that the feature k will not affect the final prediction result. Among them, the sample size in the SHAP model is a factor that affects the expected value and average prediction of SHAP. The number of samples can be reasonably controlled to achieve a more accurate explanation effect.

[0085] In one embodiment, the white box model includes decision trees, logistic regression, and decision rules.

[0086] In medical recommendation systems, the most common white-box models that can recommend and explain feature information are logistic regression models and decision rules, which themselves show the importance of each feature in the classification model to a certain recommendation result;

[0087] Another interpretable method is the feature importance of decision trees. The interpretability range of this method is called regression models. Most models using decision trees can be interpreted, and the range is relatively wide. Figure 7 , Figure 8 As shown, Figure 7 This is a diagram explaining the features of the decision tree for heart disease. Figure 8 The diagram for explaining the features of the diabetes decision tree shows that the decision tree method can explain the support vector machine, random forest and gradient boosting tree models, and the results of their prediction contribution features are roughly the same, which can be used to explain the reasons for the recommendation results and provide patients with reasons for the recommendation results.

[0088] It should be understood that, although the various steps in the above-mentioned flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above-mentioned flow chart may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0089] In one embodiment, Fig. 9 As shown, a medical decision interpretation system is provided, including: a data collection module 910, a data processing module 920, a medical decision module 930 and a decision interpretation module 940, wherein:

[0090] Data collection module 910, for collecting patient data through databases and questionnaires;

[0091] The data processing module 920 is used to perform data cleaning and data standardization on the patient data to obtain processed patient data;

[0092] The medical decision module 930 is used to input the processed patient data into the classification model, extract the data features through the classification model, and make medical prediction recommendations based on the data features; wherein logistic regression, support vector machine, random forest, gradient boosting tree, artificial neural network, k-nearest neighbor, and Bayesian network are selected as classification models;

[0093] The decision explanation module 940 is used to use the SHAP model and the white box model to explain the classification model respectively, and combine the explanation results for visual comparison, analyze the combined data features, and obtain the final decision explanation result.

[0094] like Fig.10 As shown, in one embodiment, a medical decision interpretation system is provided that further includes:

[0095] The data set division module 950 is used to randomly divide the patient data into a training set, a validation set, and a test set, and perform data cleaning and data standardization on the divided patient data to obtain processed training set, validation set, and test set data;

[0096] The model training module 960 is used to map the training set data to the initial classification model, train the initial classification model through the Nadam algorithm, change the training set data, adjust the model parameters of the initial classification model, and obtain the classification model.

[0097] In one embodiment, the model training module 960 is also used to calculate the gradient of the current parameter point in the initial classification model; use exponential moving average to calculate the first-order moment estimate and second-order moment estimate of the gradient; correct the first-order moment estimate and second-order moment estimate respectively to eliminate the deviation; use the corrected first-order moment estimate and second-order moment estimate to adjust the model parameters of the initial classification model.

[0098] In one embodiment, the model training module 960 is also used to obtain an evaluation data set and randomly divide the evaluation data set into several data subsets; select one of the several data subsets as a test subset and the rest as training subsets; train and evaluate the classification model through the test subset and the training subset, and record the accuracy, precision, and recall rate of the training and evaluation; calculate the average performance index of the classification model based on the accuracy, precision, and recall rate, and use the average performance index as the evaluation result of the classification model.

[0099] In one embodiment, the data processing module 920 is also used to perform missing value processing, outlier detection and processing, and duplicate item processing on the patient data to obtain processed preliminary data; and use text vector natural language processing technology to clean the text data in the processed preliminary data to obtain processed patient data.

[0100] In one embodiment, the data processing module 920 is also used to perform data standardization processing on the patient data using batch standardization technology in deep learning.

[0101] In one embodiment, the decision interpretation module 940 is also used to input the processed patient data into the classification model, extract data features through the classification model, and use the SHAP model to assign a SHAP value to each feature of each data point; sort each feature according to the SHAP value to achieve an interpretation analysis of the classification model; wherein the SHAP value is used to represent the contribution value of the feature.

[0102] In one embodiment, the white box model includes decision trees, logistic regression, and decision rules.

[0103] In one embodiment, a medical decision interpretation system is provided, which is operated as follows: Fig.11 As shown in the figure, after data collection and data processing, the data are divided into training set, validation set and test set. The classification model is trained by the Nadam algorithm, and the performance and prediction results of the classification model are evaluated using the k-fold cross-validation technique. The SHAP model and the white box model are used for joint interpretation and analysis.

[0104] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.12 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a medical decision interpretation method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0105] Those skilled in the art will understand that Fig.12The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0106] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps of the medical decision interpretation method when executing the computer program.

[0107] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the medical decision interpretation method are implemented.

[0108] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0109] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0110] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A medical decision interpretation method, characterized in that: The method comprises: Collect patient data through database and questionnaire; Performing data cleaning and data standardization on the patient data to obtain processed patient data; Inputting the processed patient data into a classification model, extracting data features through the classification model, and making medical prediction recommendations based on the data features; wherein logistic regression, support vector machine, random forest, gradient boosting tree, artificial neural network, k-nearest neighbor, and Bayesian network are selected as classification models; The classification models are explained respectively using the SHAP model and the white box model, and the explanation results are combined for visual comparison, and the data features are combined for analysis to obtain the final decision explanation result; The classification model is explained using the SHAP model, including: inputting the processed patient data into the classification model, extracting data features through the classification model, and assigning a SHAP value to each feature of each data point using the SHAP model; sorting each feature according to the SHAP value to achieve an explanation and analysis of the classification model; wherein the SHAP value is used to represent the contribution value of the feature.

2. The medical decision interpretation method according to claim 1, characterized in that: The method further comprises: Randomly dividing the patient data into a training set, a validation set, and a test set, and performing data cleaning and data standardization on the divided patient data to obtain processed training set, validation set, and test set data; The training process of the classification model includes: The training set data is mapped to an initial classification model, the initial classification model is trained by a Nadam algorithm, the training set data is changed, the model parameters of the initial classification model are adjusted, and a classification model is obtained; and the classification model is evaluated by using a k-fold cross validation technique.

3. The medical decision interpretation method according to claim 2, characterized in that: The step of training the initial classification model by using the Nadam algorithm and adjusting the model parameters of the initial classification model includes: Calculating the gradient of the current parameter point in the initial classification model; Calculate the first-order moment estimate and the second-order moment estimate of the gradient using exponential moving average; Correcting the first-order moment estimation and the second-order moment estimation respectively to eliminate deviations; The model parameters of the initial classification model are adjusted using the corrected first-order moment estimate and second-order moment estimate.

4. The medical decision interpretation method according to claim 2, characterized in that: The use of the k-fold cross validation technique to evaluate the classification model includes: Acquire an evaluation data set, and randomly divide the evaluation data set into a plurality of data subsets; Select one from the plurality of data subsets as a test subset and the rest as training subsets; Training and evaluating the classification model using the test subset and the training subset, and recording the accuracy, precision, and recall of the training and evaluation; The average value of the performance index of the classification model is calculated according to the accuracy, precision and recall rate, and the average value of the performance index is used as the evaluation result of the classification model.

5. The medical decision interpretation method according to claim 1, characterized in that: The data cleaning process of the patient data includes: The patient data are processed for missing values, detected and processed for outliers, and processed for duplicates to obtain processed preliminary data; The text data in the processed preliminary data is cleaned by using text vector natural language processing technology to obtain processed patient data.

6. The medical decision interpretation method according to claim 1, characterized in that: The step of performing data standardization on the patient data comprises: The batch normalization technology in deep learning is used to perform data normalization on the patient data.

7. The medical decision interpretation method according to claim 1, characterized in that: The white box model includes decision trees, logistic regression and decision rules.

8. A medical decision explanation system, characterized in that: The system comprises: Data collection module, used to collect patient data through database and questionnaire; A data processing module, used for performing data cleaning and data standardization on the patient data to obtain processed patient data; A medical decision-making module, used for inputting the processed patient data into a classification model, extracting data features through the classification model, and making medical prediction recommendations based on the data features; wherein logistic regression, support vector machine, random forest, gradient boosting tree, artificial neural network, k-nearest neighbor, and Bayesian network are selected as classification models; A decision explanation module is used to use the SHAP model and the white box model to explain the classification model respectively, and combine the explanation results for visual comparison, and analyze the data features to obtain the final decision explanation result; The decision interpretation module is also used to: input the processed patient data into the classification model, extract data features through the classification model, and use the SHAP model to assign a SHAP value to each feature of each data point; sort each feature according to the SHAP value to achieve an interpretation analysis of the classification model; wherein the SHAP value is used to represent the contribution value of the feature.

9. The medical decision interpretation system according to claim 8, characterized in that: The system further comprises: A data set division module is used to randomly divide the patient data into a training set, a validation set, and a test set, and to perform data cleaning and data standardization on the divided patient data to obtain processed training set, validation set, and test set data; The model training module is used to map the training set data to the initial classification model, train the initial classification model through the Nadam algorithm, change the training set data, adjust the model parameters of the initial classification model, and obtain the classification model; and evaluate the classification model using the k-fold cross-validation technology.

Images