Result interpretation method and device of intelligent medical diagnosis system, storage medium and computer program product

Abstract

The invention discloses a result explanation method and device of an intelligent medical diagnosis system, a storage medium and a computer program product, and relates to the technical field of intelligent medical treatment, and the method comprises the steps: obtaining clinical data of a patient; inputting the clinical data of the patient into a pre-constructed cost-sensitive decision tree regularization neural network model; and performing prediction processing on the patient clinical data through a neural network part in the cost-sensitive decision tree regularization neural network model to obtain a prediction result, and performing interpretation processing on the patient clinical data and the prediction result through a cost-sensitive decision tree regularization part to obtain an interpretation result. By introducing a cost-sensitive concept, constructing a cost-sensitive decision tree and adding the cost-sensitive decision tree into a regularization neural network model, the generalization ability of the model is improved, the intelligent medical diagnosis system keeps stable interpretation performance, and the defect that an existing intelligent medical diagnosis system lacks interpretation ability is overcome.

Description

Technical Field

[0001] The present application relates to the field of intelligent medical technology, and in particular to a result interpretation method, device, storage medium and computer program product of an intelligent medical diagnosis system. Background Art

[0002] With the development of computer hardware, computer technology, led by machine learning, has demonstrated powerful data processing capabilities. Among them, artificial intelligence methods represented by deep learning have brought new changes to the medical industry. Intelligent medical diagnosis systems combined with artificial intelligence methods such as deep learning can assist doctors in diagnosis. For example, intelligent medical diagnosis systems with powerful data processing capabilities can assist in the prediction of preeclampsia (PE), which has an ambiguous pathogenesis and a high mortality rate. However, due to the general lack of explanatory power of current intelligent medical diagnosis systems, intelligent medical diagnosis systems do not explain the diagnosis results from the perspective of human expert decision-making or do not have the ability to professionally explain the diagnosis results at all, resulting in the intelligent medical diagnosis system's auxiliary diagnosis prediction results not being accepted and trusted.

[0003] Therefore, how to make up for the lack of explanatory power of intelligent medical diagnostic systems has become a technical problem that needs to be solved urgently.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a result interpretation method, device, storage medium and computer program product of an intelligent medical diagnosis system, aiming to solve the technical problem of compensating for the lack of interpretation ability of the intelligent medical diagnosis system.

[0006] To achieve the above objectives, the present application proposes a result interpretation method of an intelligent medical diagnosis system, the method comprising:

[0007] Obtain patient clinical data;

[0008] Inputting the patient clinical data into a pre-built cost-sensitive decision tree regularized neural network model;

[0009] The neural network part in the cost-sensitive decision tree regularized neural network model performs prediction processing on the patient's clinical data to obtain a prediction result, and the cost-sensitive decision tree regularization part performs interpretation processing on the patient's clinical data and the prediction result to obtain an interpretation result.

[0010] In one embodiment, the step of inputting the patient clinical data into a pre-constructed cost-sensitive decision tree regularized neural network model, performing prediction processing through a neural network part in the cost-sensitive decision tree regularized neural network model to obtain a prediction result, and the step of the cost-sensitive decision tree regularized part interpreting the patient clinical data and the prediction result to obtain an interpretation result further includes:

[0011] Building a tree regularized neural network model based on a pre-collected training dataset;

[0012] Constructing a cost-sensitive decision tree based on the training data set and the tree regularized neural network model;

[0013] The cost-sensitive decision tree is added to the tree regularized neural network model to obtain a cost-sensitive decision tree regularized neural network model.

[0014] In one embodiment, the step of constructing a tree regularized neural network model based on a pre-collected training data set includes:

[0015] Obtain pre-collected training datasets and deep learning model frameworks;

[0016] Inputting the training data set into the deep learning model framework to obtain a neural network prediction result;

[0017] Constructing a binary classification decision tree according to the neural network prediction result and the training data set;

[0018] A regularization term is calculated based on the binary classification decision tree and the deep learning model framework;

[0019] The regularization term is added to the deep learning model for regularization processing to obtain a tree regularized neural network model.

[0020] In one embodiment, after the step of adding the regularization term to the deep learning model for regularization processing to obtain a tree regularized neural network model, the step further includes:

[0021] When the system has not reached the iteration upper limit, the training data is input into the tree regularized neural network model to obtain a new neural network prediction result, and the execution step is returned to: a binary classification decision tree is constructed according to the neural network prediction result and the training data set.

[0022] In one embodiment, the step of constructing a cost-sensitive decision tree based on the training data set and the tree regularized neural network model includes:

[0023] Obtaining feature contribution based on the training data set and the tree regularized neural network model;

[0024] A cost-sensitive decision tree is constructed according to the feature contribution.

[0025] In one embodiment, the step of adding the cost-sensitive decision tree to the tree regularized neural network model to obtain the cost-sensitive decision tree regularized neural network model includes:

[0026] Obtaining a tree regularization term based on calculating the cost-sensitive decision tree;

[0027] The tree regularization term is added to the tree regularized neural network model to obtain a cost-sensitive decision tree regularized neural network model.

[0028] In one embodiment, the step of adding the tree regularization term to the tree regularized neural network model to obtain the cost-sensitive decision tree regularized neural network model further includes:

[0029] When the system does not reach the iteration upper limit, return to the execution step: obtain the feature contribution based on the training data set and the tree regularized neural network model.

[0030] In addition, to achieve the above-mentioned objectives, the present application also proposes a result interpretation device for an intelligent medical diagnostic system, the device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the result interpretation method for the intelligent medical diagnostic system as described above.

[0031] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the result interpretation method of the intelligent medical diagnostic system as described above are implemented.

[0032] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the result interpretation method of the intelligent medical diagnosis system as described above.

[0033] One or more technical solutions proposed in this application have at least the following technical effects:

[0034] By inputting the patient's clinical data into the pre-built cost-sensitive decision tree regularized neural network model for prediction and interpretation processing, not only the prediction results are obtained, but also the interpretation results that explain the prediction results from the perspective of medical experts. By configuring the cost-sensitive decision tree regularized neural network model for the intelligent medical diagnosis system, the intelligent medical diagnosis system can not only generate prediction results, but also explain the prediction results in detail, revealing the key factors affecting the diagnosis results and the interactions between the factors. By introducing the concept of cost sensitivity, the interpretation results are consistent with the expert experience, thereby further improving the interpretation ability of the intelligent medical diagnosis system. Due to the application of regularization technology in the cost-sensitive decision tree regularized neural network model, the overfitting phenomenon of the model on the training data is reduced, the generalization ability of the model on unseen data is improved, and the intelligent medical diagnosis system maintains a stable interpretation performance, which makes up for the disadvantage of the lack of interpretation ability of the intelligent medical diagnosis system. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0037] Figure 1 A flowchart of the first embodiment of the result interpretation method of the intelligent medical diagnosis system of the present application;

[0038] Figure 2 A flowchart diagram of a second embodiment of a method for interpreting results of an intelligent medical diagnosis system of the present application;

[0039] Figure 3 A flowchart diagram of a third embodiment of a method for interpreting results of an intelligent medical diagnosis system of the present application;

[0040] Figure 4 The overall training flow chart for building a tree regularized neural network model;

[0041] Figure 5 The overall training flow chart for building a cost-sensitive decision tree regularized neural network model;

[0042] Figure 6 This is a schematic diagram of the module structure of the result interpretation device of the intelligent medical diagnosis system of the embodiment of the present application;

[0043] Figure 7 Schematic diagram of the device structure of the hardware operating environment involved in the result interpretation method of the intelligent medical diagnosis system in the embodiment of the present application.

[0044] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0045] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0046] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0047] The main solution of the embodiment of the present application is: by pre-building a cost-sensitive decision tree regularized neural network model in the intelligent medical diagnosis system, the prediction results of the intelligent medical diagnosis system are explained using the cost-sensitive decision tree regularized neural network model to obtain an explanation result that is consistent with the experience of medical experts and can be trusted. By adding a cost-sensitive decision tree to the tree regularized neural network model to obtain a cost-sensitive decision tree regularized neural network model, a simpler tree-structured explanation diagram that is more consistent with expert knowledge can be generated, making up for the current intelligent medical diagnosis system's lack of explanation ability.

[0048] This embodiment takes into account that for some diseases with complex definitions and ambiguous pathogenesis, relying solely on clinical features for diagnosis will increase the possibility of misdiagnosis. For example, preeclampsia (PE), PE is a pregnancy syndrome, one of the most common medical complications during pregnancy, and one of the main causes of maternal mortality. However, the clinical characteristics of PE are not obvious, and there is no effective treatment except for induced labor. The powerful data processing capabilities demonstrated by the intelligent medical diagnosis system can assist doctors in diagnosing and predicting diseases including PE. However, since most of the current intelligent medical diagnosis systems cannot explain the prediction results, there is no interpretability. The models used in the intelligent medical diagnosis system that provides interpretability only use the contribution of the calculated model features to explain the model prediction results. For example, an intelligent medical diagnosis system that uses the SHAP method to explain the prediction results, although the feature contribution of the SHAP method can represent the importance of different features of the prediction results to a certain extent, thereby explaining the focus of the model prediction, the SHAP method does not explain the model prediction process, and its explanation of the model is still based on data-driven. The interpretable methods for calculating the contribution of model features, led by the SHAP method, have not been explained from the perspective of human expert decision-making, which has led to the prediction results of the intelligent medical diagnosis system not being accepted and trusted by medical experts. This phenomenon shows that the current medical diagnosis system lacks the ability to explain and it is difficult to give a strong explanation of the prediction results, making it difficult for the prediction results to be trusted by the public.

[0049] Therefore, this application provides a solution by adding a cost-sensitive decision tree to a tree regularized neural network model to construct a cost-sensitive decision tree regularized neural network model. The cost-sensitive decision tree regularized neural network model is used to explain the prediction results of the intelligent medical diagnosis system from the perspective of human expert decision-making, so that the prediction results are accepted and trusted by experts in the medical field, making up for the current intelligent medical diagnosis system's lack of explanatory power.

[0050] It should be noted that the execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, an intelligent medical diagnosis system, etc. The following takes the intelligent medical diagnosis system as an example to illustrate this embodiment and the following embodiments.

[0051] Based on this, the present application embodiment provides a result interpretation method of an intelligent medical diagnosis system, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the result interpretation method of the intelligent medical diagnosis system of the present application.

[0052] In this embodiment, the result interpretation method of the intelligent medical diagnosis system includes steps S10 to S20:

[0053] Step S10, obtaining patient clinical data;

[0054] Intelligent medical diagnosis system is a medical auxiliary tool that uses artificial intelligence technology to quickly and accurately diagnose patients' diseases and provide treatment recommendations. Intelligent medical diagnosis system is usually based on big data analysis, machine learning and deep learning technologies, which can mine correlations and regularities from massive amounts of medical clinical data and establish the relationship between diseases and symptoms and signs.

[0055] Acquiring clinical data of patients is a prerequisite for the intelligent medical diagnosis system to generate prediction results and corresponding interpretation results. Sensors can be used to monitor the patient's physiological parameters, disease indicators, etc. in real time. Sensors can send the collected data to the cloud via wireless transmission technology for further analysis and prediction by the intelligent medical diagnosis system.

[0056] Step S20, inputting the patient clinical data into a pre-built cost-sensitive decision tree regularized neural network model;

[0057] The cost-sensitive decision tree is a decision tree algorithm that considers the misclassification costs of different categories. It adjusts the tree construction process by introducing a cost matrix so that the tree can consider the misclassification costs of different categories when classifying. By adding doctor expert knowledge as cost, the model prioritizes the features that doctors consider important while ensuring the prediction results, and generates interpretation results that are consistent with doctors' experience.

[0058] The tree regularized neural network model is a model that combines decision trees and regularized neural networks. Regularized neural networks are models that introduce regularization terms into neural networks to prevent overfitting. Regularization terms usually include L1 regularization and L2 regularization, which penalize the complexity of the model by adding the sum of the absolute values ​​of weights or the sum of the squares of weights to the loss function, encouraging the model to learn simpler and more generalized features. The tree regularized neural network model not only combines the intuitive interpretability of decision trees and the powerful learning ability of neural networks, but also uses regularization to prevent overfitting.

[0059] The cost-sensitive decision tree regularized neural network model further introduces the concept of cost-sensitive decision tree based on the tree regularized neural network model. This means that when building the model, not only regularization is considered to prevent overfitting, but also the misclassification costs of different categories are considered to optimize the classification performance.

[0060] It is understandable that although the tree regularized neural network model is a model that combines a decision tree and a regularized neural network, the decision tree in the model is just an ordinary decision tree algorithm and does not consider the misclassification costs of different categories. It is necessary to further add a cost-sensitive decision tree algorithm to the model and introduce expert knowledge in the medical field as the cost. The final model can explain the prediction results from the perspective of medical experts, so that the prediction results can be accepted and trusted, which can make up for the current intelligent medical diagnosis system's lack of explanatory power.

[0061] Step S30, the neural network part of the cost-sensitive decision tree regularized neural network model performs prediction processing on the patient clinical data to obtain a prediction result, and the cost-sensitive decision tree regularized part performs interpretation processing on the patient clinical data and the prediction result to obtain an interpretation result

[0062] Prediction results refer to the automated output results of the patient's disease condition, risk prediction or treatment recommendations given by the intelligent medical diagnosis system based on the input medical data after a series of technical analyses. The prediction results include the type of disease, degree of lesions, development trend, treatment recommendations, etc., and are intended to assist doctors in making more accurate diagnoses and treatment decisions.

[0063] The patient's clinical data is used as input data and passed to the cost-sensitive decision tree regularized neural network model. The neural network model part of the cost-sensitive decision tree regularized neural network model analyzes the patient's clinical data to obtain the prediction results. The cost-sensitive decision tree regularized part of the cost-sensitive decision tree regularized neural network model interprets the prediction results to obtain the interpretation results that are consistent with the decision-making perspective of medical experts.

[0064] Specifically, as a feasible implementation, step S30 may include: first, passing the patient's clinical data as input data to the tree regularized neural network model part of the cost-sensitive decision tree regularized neural network model to obtain a prediction result. Furthermore, the cost-sensitive decision tree regularized neural network model uses a cost-sensitive decision tree to analyze the prediction result. According to a predefined cost matrix, the misclassification costs of different categories are considered, the most appropriate classification path is selected, and an explanation result is generated. The explanation result is a tree-structured explanation diagram, which usually involves an explanation of the model decision process, such as which features have an important impact on the classification results, and the prediction probabilities of different categories. These explanation results can help doctors better understand the decision-making basis of the model and improve the reliability and interpretability of the prediction results.

[0065] The above is only a feasible implementation of step S30 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S30.

[0066] This embodiment provides a method for interpreting the results of an intelligent medical diagnosis system. By inputting the patient's clinical data into a pre-built cost-sensitive decision tree regularized neural network model for prediction processing and interpretation processing, not only the prediction results are obtained, but also the interpretation results that interpret the prediction results from the perspective of medical experts. By configuring a cost-sensitive decision tree regularized neural network model for the intelligent medical diagnosis system, the intelligent medical diagnosis system can not only generate prediction results, but also interpret the prediction results in detail, revealing the key factors affecting the diagnosis results and the interactions between the factors. By introducing the concept of cost sensitivity, the interpretation results are consistent with expert experience, thereby further improving the interpretation ability of the intelligent medical diagnosis system. Due to the application of regularization technology in the cost-sensitive decision tree regularized neural network model, the overfitting phenomenon of the model on the training data is reduced, the generalization ability of the model on unseen data is improved, and the intelligent medical diagnosis system maintains a stable interpretation performance, making up for the disadvantage of the lack of interpretation ability of the intelligent medical diagnosis system.

[0067] Based on the first embodiment of the present application, the second embodiment of the present application is proposed. In the second embodiment of the present application, the same or similar contents as those of the above-mentioned first embodiment can be referred to the above introduction, and will not be repeated in the following.

[0068] On this basis, please refer to Figure 2 , Figure 2 This is a flow chart of the second embodiment of the result interpretation method of the intelligent medical diagnosis system of the present application.

[0069] In this embodiment, before step S20 of inputting the patient clinical data into the pre-built cost-sensitive decision tree regularized neural network model, steps S121 to S123 are also included:

[0070] Step S121, constructing a tree regularized neural network model based on the pre-collected training data set;

[0071] The pre-collected training datasets usually contain a lot of historical clinical data with a lot of information and features. By building a tree-regularized neural network model, the complex structures and patterns in these data can be effectively utilized. The tree-regularized neural network model can learn the deep relationships in the data and continuously optimize its own structure during the training process to better adapt to the characteristics of the data.

[0072] The tree regularized neural network model refers to the use of a decision tree structure as a regularization method to improve the generalization ability of the neural network model and prevent the model from overfitting.

[0073] Building a tree regularized neural network model based on the pre-collected training data set can fully utilize the characteristics and advantages of the data and improve the generalization ability and performance of the model. It is of great significance for dealing with complex medical problems and can assist doctors in making more accurate and reliable decisions.

[0074] Specifically, in a feasible implementation manner, step S121 may include:

[0075] First, prepare the data needed to build the tree regularized neural network model.

[0076] Preprocess the pre-collected training data set, including data cleaning, missing value processing, feature selection or extraction, and data normalization, to ensure that the quality and format of the data set are suitable for model training.

[0077] Furthermore, a basic neural network structure is designed to construct a tree regularized neural network model.

[0078] Design a basic neural network structure, including determining the number of layers, the number of neurons in each layer, and the type of activation function. Choose a network architecture suitable for the task, such as a fully connected network, a convolutional neural network, or a recurrent neural network. The choice of the specific neural network structure depends on the type of data and the nature of the problem.

[0079] Furthermore, a tree regularization term is introduced in the training process of the neural network.

[0080] Tree regularization is usually achieved by adding a regularization term related to the neural network weights to the loss function. The regularization term can be defined based on certain characteristics of the decision tree, such as the splitting criterion of the decision tree or the complexity of the tree. For example, the splitting criterion of the decision tree is transformed into a constraint on the neural network weights. By defining a regularization term, the network weights are encouraged to produce sparsity near the decision boundary, thereby introducing a tree regularization term in the neural network infrastructure. Another example: the introduction of the tree regularization term is achieved by combining the output of the neural network with the structure of the decision tree. The neural network is responsible for learning complex feature representations, and the decision tree performs the final classification or regression based on these representations.

[0081] Furthermore, the trained tree regularized neural network model is verified and evaluated.

[0082] Use the validation data set to evaluate the trained model and check whether its performance meets expectations. Based on the validation results, the model structure, parameters, or regularization terms can be adjusted to optimize the model performance.

[0083] Finally, the performance of the model is evaluated on the test dataset to ensure that the model has good generalization ability. Once the model reaches a satisfactory performance level, it can be applied in real-world scenarios.

[0084] It should be noted that the specific implementation details and parameter settings need to be adjusted and optimized according to the specific tasks and data sets. In addition, due to the complexity of the model and the high computational cost of the training process, it is necessary to use high-performance computing resources or distributed computing frameworks to accelerate the model training and evaluation process.

[0085] In this embodiment, a suitable neural network model framework is constructed based on the pre-collected data set, the neural network model framework is trained using the data set, and the trained model is verified and evaluated to ensure that the trained model has good generalization ability and prevent overfitting. It can realize the interpretation of the prediction results, so that the intelligent medical diagnosis system has the ability to interpret the prediction results.

[0086] The above is only a feasible implementation of step S121 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S121.

[0087] Step S122, constructing a cost-sensitive decision tree based on the training data set and the tree regularized neural network model;

[0088] Although the tree regularized neural network model has the ability to explain the prediction results, its explanation results are not consistent with expert experience. By adding cost-sensitive operations to the original tree regularized neural network to construct a cost-sensitive decision tree, it is possible to explain the prediction results from the perspective of medical experts.

[0089] The construction of a cost-sensitive decision tree requires defining a cost matrix based on the training data set, introducing the experience and knowledge of medical experts into the cost matrix as cost, and constructing a cost-sensitive decision tree based on the cost matrix and the feature representation extracted from the tree regularized neural network model.

[0090] First, a regularized neural network model is used to extract feature representations from the training dataset. The training data is input into the model, and the output of the hidden layer or the last layer of the model is obtained as the feature representation. These feature representations capture the deep structure and patterns of the data, which are crucial for the subsequent construction of a cost-sensitive decision tree.

[0091] Furthermore, a cost matrix is ​​defined to quantify the cost of misclassification of different categories. The cost matrix is ​​a two-dimensional array in which each element represents the cost of misclassifying a sample from one category to another. This matrix reflects the relative importance of misclassification of different categories and can be set according to the actual application scenario.

[0092] Furthermore, a cost-sensitive decision tree is constructed based on the feature representation and the cost matrix. This process is similar to the process of constructing a standard decision tree, but the misclassification cost in the cost matrix needs to be considered. At each node of the tree, the impact of different feature partitions on the cost is evaluated, and the partition that minimizes the total cost is selected. The best partition is selected by calculating the cost-weighted purity improvement of each partition.

[0093] Furthermore, in order to prevent overfitting and improve the generalization ability of the cost-sensitive decision tree, it is necessary to prune the constructed cost-sensitive decision tree. The pruning process includes deleting some subtrees or leaf nodes to simplify the tree structure and reduce the risk of overfitting. In addition, other optimization operations can be performed on the tree, including adjusting the depth of the tree, the minimum number of samples of leaf nodes, and other parameters.

[0094] Finally, the constructed cost-sensitive decision tree is evaluated using the validation data set. By calculating the performance indicators of the model on the validation data set (such as accuracy, cost loss, etc.), the effectiveness and generalization ability of the model can be evaluated. Based on the evaluation results, the model can be further adjusted and optimized.

[0095] A cost-sensitive decision tree that takes misclassification into account is constructed based on the training data set and the tree regularized neural network model. By introducing expert knowledge, the prediction results are explained from the perspective of expert decision-making, generating a simpler tree structure explanation diagram that is more in line with expert knowledge, so that the prediction results can be recognized and trusted.

[0096] Step S123, adding the cost-sensitive decision tree to the tree regularized neural network model to obtain a cost-sensitive decision tree regularized neural network model.

[0097] Adding the cost-sensitive decision tree to the tree regularized neural network model actually combines the advantages of the decision tree algorithm and the neural network algorithm to build a cost-sensitive decision tree regularized neural network model. The cost-sensitive decision tree regularized neural network model is a high-level conceptual description that can explain the prediction results from the perspective of expert decision-making, making up for the lack of explanatory power of the intelligent medical diagnosis system.

[0098] Adding a cost-sensitive decision tree to a tree-regularized neural network model requires modifying the loss function of the tree-regularized neural network model. Using the cost-sensitive decision tree as inspiration for the tree-regularized neural network layer, the structure of the tree-regularized neural network model is simplified through a pruning strategy. Furthermore, a neural network with a cost-sensitive loss function is used for training while applying a tree regularization strategy. Finally, multiple hyperparameters such as learning rate, regularization strength, and cost-sensitive weights are adjusted to find the best model configuration.

[0099] It should be noted that the cost-sensitive decision tree regularized neural network model combines the advantages of two different model frameworks, which may make the training of the cost-sensitive decision tree regularized neural network model more complicated and difficult to train. Therefore, in practical applications, sufficient experiments and verification are required to ensure the effectiveness and practicality of the proposed model.

[0100] In this embodiment, a tree regularized neural network model is constructed by using a pre-collected training data set, so that the intelligent medical diagnosis system has the ability to explain the prediction results. In order to enable the intelligent medical diagnosis system to explain the prediction results from the perspective of medical experts and completely make up for the current medical diagnosis system's lack of explanatory power, a cost-sensitive decision tree is constructed and added to the tree regularized neural network model to obtain the final cost-sensitive decision tree regularized neural network model. This model can analyze the prediction results from the perspective of expert decision-making, so that the prediction results of the intelligent medical diagnosis system can be trusted and recognized, thereby improving the use value of the prediction results.

[0101] Based on the first embodiment and / or the second embodiment of the present application, the third embodiment of the present application is proposed. In the third embodiment of the present application, the same or similar contents as those of the above-mentioned first and second embodiments can be referred to the above introduction, and will not be repeated in the following.

[0102] On this basis, please refer to Figure 3 , Figure 3 A flowchart of a third embodiment of the data cleaning method provided in the present application.

[0103] In this embodiment, step S121 of constructing a tree regularized neural network model according to the pre-collected training data set includes steps S1 to S5:

[0104] Step S1, obtaining a pre-collected training data set and a deep learning model framework;

[0105] In order to ensure the learning effect and performance of the final model and improve the efficiency of training and deployment, the deep learning model framework is selected as the basis for training the model. The pre-collected training data set is usually pre-processed and labeled, making the model training process more efficient.

[0106] The deep learning model framework has multiple hidden layers and can learn and represent the complexity of data. The multilayer perceptron (MLP) has an input layer, one or more hidden layers, and an output layer. It is the simplest model framework. Since the number of network layers of the deep learning model framework is much larger than that of the multilayer perceptron, the main difference between the two is the number of network layers. For ease of understanding, we will take the simplest multilayer perceptron MLP as an example to explain how to obtain a training data set. Suppose a single sample input is x n , the corresponding true label is represented as y n ,The training data set is a combination of sample input and the corresponding ,true label including the real feature data of the sample, such as the clinical data of the ,patient, such as height, weight, blood indicators and so on.

[0107] Step S2, inputting the training data set into the deep learning model framework to obtain a neural network prediction result;

[0108] Input the training data set into the deep learning model framework to obtain the initial neural network prediction results. The neural network result is actually a prediction label corresponding to the sample in the pre-training data set. The deep learning model framework is used to convert the sample data in the training data set into a prediction label, that is, the neural network prediction result. The real label corresponding to the sample data in the training data set can be used to test and evaluate the constructed model.

[0109] Since the single sample input is x n , the corresponding true label is represented as y n , the prediction function in the deep learning model framework is expressed as in Represents the prediction result, that is, the neural network prediction result, sample data x n The predicted label of , and the vector W is the model framework parameter.

[0110] Step S3, constructing a binary classification decision tree according to the neural network prediction result and the training data set;

[0111] The neural network prediction results are used as labels and the training data set is used as input to build a binary classification decision tree. In fact, the neural network prediction results output by the deep learning model framework are used as the training target of the binary classification decision tree, and the binary classification decision tree is built based on the characteristics of the training data set.

[0112] According to the prediction function in the deep learning model framework Using the Sample Collection and the corresponding neural network net(W) prediction set Construct a proxy dataset and input it into a basic decision tree model framework to generate a binary classification decision tree that can accurately reproduce the given sample input and network prediction.

[0113] Step S4, calculating a regularization term based on the binary classification decision tree and the deep learning model framework;

[0114] The average path length of the generated binary decision tree is calculated, and the average path length is used as the prediction target. The parameters in the deep learning model framework are used as input and input into an independent MLP network for fitting. The MLP calculates the estimated value of the average path length based on the parameters of the deep learning model framework and outputs the estimated value as a regularization term.

[0115] First, calculate the average path length of the generated binary classification decision tree.

[0116] The average path length is defined as n The average number of decision tree nodes that must be passed when making a prediction. Assuming that the function for calculating path length is Path Len(tree, x), the calculation formula for average path length is as follows:

[0117]

[0118] In the calculation formula of the average path length, the average path length is used as the output of the tree regularization function Ω(W), N represents the data capacity of the samples included in the training data set, and n=1.

[0119] Furthermore, the average path length is used as the prediction target and the parameters in the deep learning model framework are used as input to an independent MLP network for fitting. The MLP calculates the estimated value of the average path length based on the parameters of the deep learning model framework and outputs the estimated value as a regularization term.

[0120] According to the calculation formula of the average path length above, the complete algorithm of the complete tree regularization function Ω(W) is as follows:

[0121]

[0122] It should be noted that Ψ(W) represents the tree regularization term and λ represents the strength of the regularization term. The input is a dataset containing N samples. And predictions involving neural networks The output is the average path length.

[0123] In addition, it should be noted that the tree regularization function Ω(W) is not differentiable. In the implementation steps of the tree regularization function, the process of building a decision tree is not differentiable, which leads to the tree regularization function Ω(W) being also not differentiable with respect to the neural network parameter W. This makes it impossible to apply gradient descent to iterative training of the neural network after adding the tree regularization term. Although derivative-free optimization techniques can be used, it will greatly increase the training cost.

[0124] Therefore, we choose to introduce the proxy function Instead of the original tree regularization function Ω(W), the tree regularization term is calculated. This proxy function is implemented by an independent MLP, and the output of the tree regularization function Ω(W) is obtained by mapping the original neural network parameters W, that is, the estimated value of the average path length. It is implemented through an independent MLP and is differentiable relative to the original neural network.

[0125] Let the vector θ represent the parameters of the selected independent MLP network approximator, ∈ is the regularization strength, whose value is greater than 0, then the proxy function The optimization can be trained by minimizing the squared error loss with the tree regularization function Ω(W), as shown in the following formula:

[0126]

[0127] It should be noted that the vector θ represents the parameters of the selected independent MLP network approximator, ∈ is the regularization strength, is the proxy function, Ω(W) is the tree regularization function, and j=1.

[0128] The above formula can achieve the effect of training optimization, so that the estimated value of the average path length reaches the optimal solution, providing a basis for subsequent model construction.

[0129] Step S5, adding the regularization term to the deep learning model for regularization processing to obtain a tree regularized neural network model.

[0130] By integrating the calculation logic of the loss function and the tree regularization term into the deep learning model framework, the model framework is trained in a loop so that the regularization term is calculated and applied in each iteration, thereby obtaining a tree regularized neural network model.

[0131] In the process of training the model framework, it involves passing the input data to the model and obtaining the output of the model. Furthermore, a custom loss function is used to calculate the loss between the output and the true label, including data fitting loss and tree regularization terms. Furthermore, the gradient is calculated based on the loss, and the gradient is propagated back to the parameters of the model. Finally, the optimizer is used to update the parameters of the model based on the gradient, thereby realizing the construction of the tree regularized neural network model.

[0132] In this embodiment, the training data set is input into the deep learning model framework to obtain the neural network prediction result; the neural network prediction is used as a label and the training data is used as input to construct a binary classification decision tree; the average path length of the generated binary classification decision tree is calculated; the average path length is used as the prediction target and the parameters of the deep learning model are used as input, and input into an independent MLP network for fitting; the MLP calculates the estimated value of the average path length based on the parameters of the deep learning model and outputs it as a regularization term; the regularization term is added to the deep learning model and regularized to obtain a tree regularized neural network model. By configuring the tree regularized neural network model in the intelligent medical diagnosis system, explainability is added to the prediction result, so that the intelligent medical diagnosis system has a certain ability to explain the prediction result.

[0133] Based on the above embodiments of the present application, a fourth embodiment of the present application is proposed. In the fourth embodiment of the present application, the same or similar contents as those in the above embodiments can be referred to the above introduction, and will not be described in detail later.

[0134] In this embodiment, the regularization term is added to the deep learning model for regularization processing, and step S5 to obtain a tree regularized neural network model further includes step S6:

[0135] When the system has not reached the iteration upper limit, the training data is input into the tree regularized neural network model to obtain a new neural network prediction result, and the execution step is returned to: a binary classification decision tree is constructed according to the neural network prediction result and the training data set.

[0136] The parameters of the tree regularized neural network model, such as weights and biases, are randomly initialized or set through some pre-training method. Through iterative training, these initial parameters can be adjusted according to the results of the previous round of training, so that the model can better fit the data and improve the prediction performance.

[0137] It is understandable that during the first data training process, the training data set is input into the deep learning model framework to obtain the neural network prediction results. After a round of data training, the deep learning model framework incorporates the tree regularization term to become a tree regularized neural network model with regularization capability. In order to better fit the data and improve the prediction performance, the tree regularized neural network model after the first round of data training is iteratively trained.

[0138] When the system has not reached the preset iteration limit, the training data set is input into the tree regularized neural network model constructed in the previous iteration process, and a new neural network prediction result is obtained according to the newly constructed tree regularized neural network model.

[0139] After obtaining the new neural network prediction results, return to the execution step: construct a binary classification decision tree based on the neural network prediction results and the training data set, which is consistent with the steps and calculation formulas for constructing a binary classification decision tree in the previous round. After constructing the new binary classification decision tree, continue to calculate the average path length of the binary classification decision tree, and iteratively train the tree regularization neural network model based on the average path length to improve the generalization ability of the final model and prevent overfitting.

[0140] In this embodiment, the tree regularized neural network model is iteratively trained, and before the system reaches a preset number of iterations, iterative training is performed based on the model building results of the previous round. The model parameters are continuously adjusted and the loss function is optimized so that the model can achieve better performance on a given task.

[0141] Based on the above embodiments of the present application, a fifth embodiment of the present application is proposed. In the fifth embodiment of the present application, the same or similar contents as those in the above embodiments can be referred to the above introduction, and will not be repeated in the following.

[0142] In this embodiment, step S122 of constructing a cost-sensitive decision tree based on the training data set and the regularized neural network model may include steps A10 to A20:

[0143] Step A10, obtaining feature contribution based on the training data set and the regularized neural network model;

[0144] The training data set is input into the regularized neural network model, and the SHAP method is used to obtain the feature contribution of each feature in the training data set, namely the Shapley value.

[0145] It should be noted that SHAP (SHapley Additive exPlanations) is a method based on the Shapley value explanation model in game theory, which is used to explain the predictions of machine learning models. This method can calculate the importance value (Shapley value) of each feature variable in each sample, thereby achieving the purpose of explaining the prediction effect of the model.

[0146] SHAP is used to assign importance to different features. On the premise of ensuring the accuracy of prediction results, features that doctors consider important are given priority for modeling, and explanation results that are consistent with doctors' experience are generated.

[0147] Since the explanation results are consistent with the doctor's knowledge, it is more conducive to the doctor's acceptance of the diagnosis model and trust in the diagnosis results. Taking the medical field as an example, the collected data includes data such as the patient's height, weight, plasma protein A content, and plasma urea nitrogen content. According to the doctor's knowledge, the feature of the patient's plasma urea nitrogen content is more conducive to the explanation of the disease diagnosis results. If the explanatory model constructed using the plasma urea nitrogen content has similar performance to the explanatory model constructed using the other three features, the use of the more important plasma urea nitrogen content as the feature to construct the explanatory model is given priority.

[0148] Step A20: construct a cost-sensitive decision tree according to the feature contribution.

[0149] The cost-sensitive decision tree is constructed using the iterative binary tree 3 generation algorithm. The standard for dividing data is information gain, which is calculated based on information entropy. The calculation method of information entropy is as follows:

[0150]

[0151] In the formula, D represents the sample set, p k It represents the proportion of samples of the kth category in the sample set D, and |y| represents the number of sample categories in the sample set.

[0152] The calculation formula of information gain is as follows:

[0153]

[0154] Where D v Indicates that the feature a in D takes the value a v , Gain(D,a) represents the calculated information gain.

[0155] Cost-sensitive operations require adding the cost value corresponding to the attribute in the calculation process of information gain, so that the original information gain calculation method is converted into cost information gain. The calculation method is as follows: CostGain(D, a) = Gain(D, a) / Cost(a), where Cost(a) is the cost of attribute a, Gain(D, a) represents the calculated information gain, and CostGain(D, a) represents the cost information gain.

[0156] In order to cope with the scenario where expert knowledge cannot be added as cost, feature contribution is selected as an additional metric to measure the importance of features, achieving the same effect as cost information gain.

[0157] The cost information gain is obtained through the above calculations. Combined with the feature contribution calculated by the SHAP method, the iterative binary tree 3rd generation algorithm is used to realize the construction of a cost-sensitive decision tree.

[0158] It should be noted that the iterative binary tree three-generation algorithm includes three parts: pre-order traversal, in-order traversal and post-order traversal. It mainly uses top-down greedy search to traverse the possible decision space. Greedy search means taking the best decision at each step, hoping that such a local optimal choice can lead to a global optimal solution. The cost-sensitive decision tree constructed by this algorithm can find an explanation path that is more in line with the expert's perspective, thereby generating subsequent explanation results.

[0159] In this embodiment, by adding expert knowledge as cost, calculating cost information gain, and considering the case where expert knowledge cannot be added as cost, a cost-sensitive decision tree is constructed in combination with the feature contribution calculated using the SHAP method. By constructing a cost-sensitive decision tree, the final model can explain the prediction results of the intelligent medical diagnosis system from the perspective of expert decision-making, thereby making up for the current intelligent medical diagnosis system's lack of explanatory power.

[0160] Based on the above embodiments of the present application, the sixth embodiment of the present application is proposed. In the sixth embodiment of the present application, the same or similar contents as those in the above embodiments can be referred to the above introduction, and will not be repeated in the following.

[0161] In this embodiment, the step S123 of adding the cost-sensitive decision tree to the tree regularized neural network model to obtain the cost-sensitive decision tree regularized neural network model may include steps A30 to A40:

[0162] Step A30, obtaining a tree regularization term based on calculating the cost-sensitive decision tree;

[0163] Since the cost factor is considered in the selection of each split point during the construction of the decision tree, unlike the traditional decision tree, the cost-sensitive decision tree will give priority to those split points that can minimize the overall cost.

[0164] The purpose of the tree regularization term is to control the complexity of the model and prevent overfitting. In cost-sensitive decision trees, the calculation of the regularization term may be combined with cost factors.

[0165] Therefore, the tree regularization term is calculated by the split points generated in the process of building the cost decision tree, and the specific tree regularization term is obtained by assigning different weights to different split points. When calculating the tree regularization term, some hyperparameter adjustments may also be involved, such as regularization strength, the upper limit of the number of leaf nodes, etc. The choice of these hyperparameters will affect the balance between the performance and complexity of the model. It is necessary to combine the weighted loss function to optimize the calculation of the tree regularization term.

[0166] Step A40, adding the tree regularization term to the regularized neural network model to obtain a cost-sensitive decision tree regularized neural network model.

[0167] The loss function usually consists of two parts: one is the traditional loss term (such as cross entropy loss or mean square error loss), which is used to measure the difference between the model prediction and the true label; the other is the tree regularization term, which is used to control the complexity of the model and consider cost factors. By adding the tree regularization term to the loss function of the regularized neural network model, the construction of the cost-sensitive decision tree regularized neural network model is realized.

[0168] In this embodiment, by calculating the tree regularization term of the cost-sensitive decision tree, the tree regularization term of the cost-sensitive decision tree is added to the tree regularized neural network model to obtain the cost-sensitive decision tree regularized neural network model. Expert knowledge is introduced as cost in the cost-sensitive decision tree regularized neural network model, so that the intelligent medical diagnosis system can generate a tree-like explanation diagram, explain the prediction results from the perspective of medical experts, and make up for the current intelligent medical diagnosis system's lack of explanatory power.

[0169] Based on the above embodiments of the present application, a seventh embodiment of the present application is proposed. In the seventh embodiment of the present application, the same or similar contents as those of the above embodiments can be referred to the above introduction, and will not be repeated in the following.

[0170] In this embodiment, after step A40 of adding the tree regularization term to the regularized neural network model to obtain the cost-sensitive decision tree regularized neural network model, step A50 is also included;

[0171] Step A50, when the system has not reached the iteration upper limit, returns to the execution step: obtaining the feature contribution based on the training data set and the regularized neural network model.

[0172] The regularization term in the regularized neural network model is calculated from the parameters of the binary decision tree and the deep learning model framework. After adding the tree regularization term of the cost-sensitive decision tree, the regularization term of the regularized neural network model changes, causing the model parameters to change accordingly, and the feature contribution also changes. In order to ensure the explanatory effect of the model, it is necessary to re-obtain the feature contribution based on the adjusted cost-sensitive decision tree regularized neural network model and further optimize the model.

[0173] In order to achieve better optimization effect on the cost-sensitive decision tree regularized neural network model and adjust the model parameters, it is necessary to first set the iteration upper limit. When the system does not reach the iteration upper limit, gradually improve the model performance and find more accurate and meaningful feature contributions.

[0174] It should be noted that the iteration limit should be set reasonably during the iteration process to avoid overfitting or waste of computing resources. It is also necessary to pay close attention to the performance of the model and the changing trend of feature contribution so as to adjust the optimization strategy in time.

[0175] Furthermore, according to a pre-set iteration upper limit, the feature contribution is recalculated using the training data set and the regularized neural network model.

[0176] If the iteration limit is not reached, the regularized neural network model is adjusted according to the results of the previous iteration or new requirements. The regularized neural network model is adjusted by changing the type of regularization term, adjusting the regularization strength, modifying the network structure, etc. The feature contribution is recalculated using the training data set and the adjusted regularized neural network model.

[0177] In this embodiment, when the system has not reached the iteration limit, the cost-sensitive decision tree regularized neural network model is continuously adjusted according to the evaluation results of the current feature contribution, so that the model can more accurately capture the feature relationship in the data. Each iteration can further optimize the parameters of the model, thereby improving the interpretation performance.

[0178] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the result interpretation method of the intelligent medical diagnosis system of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0179] Reference Figure 4 As shown, Figure 4 The overall training flow chart for building a tree regularized neural network model. Figure 4 , the overall training process of constructing a tree regularized neural network model in this application is explained.

[0180] First, the training data set is input into the deep learning model framework to obtain the neural network prediction results. The deep learning model framework is a subset of the neural network, and the neural network is the general concept of the deep learning model. Figure 4 The neural network shown in the specific operation process is a deep learning model framework. The training data set includes sample data and the true label corresponding to the sample data. The neural network prediction result obtained by inputting the training data set into the deep learning model framework is actually the prediction label corresponding to the sample data.

[0181] Furthermore, the prediction results of the neural network are used as labels and the training data set is used as input to construct a binary classification decision tree. The prediction results of the neural network are combined with the sample data in the training data set into a data set, and the data set is fitted to obtain a binary classification decision tree.

[0182] Furthermore, the complexity of the binary classification decision tree is calculated. The complexity of the binary classification decision tree is obtained by calculating the average path length of the binary classification decision tree. The average path length is the average number of decision tree nodes that must be passed when predicting sample data.

[0183] Furthermore, the average path length is used as the prediction target and the parameters of the deep learning model framework are used as inputs to an independent MLP network for fitting. The average path length and the basic parameters of the deep learning model framework are combined into a proxy dataset, which is input into an independent MLP network for fitting. The MLP calculates the estimated value of the average path length based on the parameters of the deep learning model framework, and outputs the estimated value of the average path length as a regularization term.

[0184] Finally, the calculated regularization term is added to the deep learning model framework to regularize the model and obtain a tree regularized neural network model. The deep learning model framework is a type of neural network. By obtaining the existing model framework, the model training process can be shortened. The calculated regularization term is actually an estimate of the average path length of the binary decision tree, so the deep learning model with the regularization term added is named the tree regularized neural network model. In order to ensure the performance of the model and prevent the model from overfitting, the model needs to be iteratively trained.

[0185] Reference Figure 5 As shown, Figure 5 The overall training flow chart for constructing a cost-sensitive decision tree regularized neural network model is shown below. Figure 5 , the overall training process of constructing a cost-sensitive decision tree regularized neural network model in this application is explained.

[0186] First, the training data set is input into the tree regularized neural network model obtained in the previous round of model building process, and the SAHP method is applied in the tree regularized neural network model to obtain the feature contribution corresponding to each sample data in the training data set, that is, Figure 5 The Shapley value shown.

[0187] Furthermore, the feature contribution is used as the cost to construct a cost-sensitive decision tree. Expert knowledge is also introduced as the cost in the cost-sensitive decision tree. The feature contribution is used as the cost to deal with the situation where expert knowledge cannot be added as the cost, so the Shapley value is used as an additional indicator to measure the importance of the feature and expand the applicability of the model.

[0188] Furthermore, the complexity of the cost-sensitive decision tree is calculated to obtain a tree regularization term, and the tree regularization term is added to this round of regularized neural network model to obtain a cost-sensitive decision tree decision regularized neural network model.

[0189] Finally, in order to ensure the performance of the model and prevent overfitting, the cost-sensitive decision tree decision regularization neural network model is iteratively trained. When the system does not reach the iteration limit, the model continues to be trained and returns to the calculation step of feature contribution.

[0190] This application also provides a result interpretation device for an intelligent medical diagnosis system, please refer to Figure 6 , the result interpretation device of the intelligent medical diagnosis system comprises:

[0191] The prediction module 10 is used to obtain the patient's clinical data, input the patient's clinical data into a pre-created cost-sensitive decision tree regularized neural network model; perform prediction processing through the neural network part of the cost-sensitive decision tree regularized neural network model to obtain a prediction result;

[0192] Acquiring clinical data of patients is a prerequisite for the intelligent medical diagnosis system to generate prediction results and corresponding interpretation results. Sensors can be used to monitor the patient's physiological parameters, disease indicators, etc. in real time. Sensors can send the collected data to the cloud through wireless transmission technology for further analysis and prediction by the intelligent medical diagnosis system.

[0193] The patient's clinical data is used as input data and passed to the cost-sensitive decision tree regularized neural network model. The neural network model part of the cost-sensitive decision tree regularized neural network model performs predictive processing on the patient's clinical data to obtain a prediction result.

[0194] The explanation module 20 is used to interpret the patient clinical data and prediction results through the cost-sensitive decision tree regularization part in the cost-sensitive decision tree regularization neural network model to obtain an explanation result.

[0195] The cost-sensitive decision tree regularization part of the cost-sensitive decision tree regularization neural network model combines the patient's clinical data to analyze the prediction results and obtain an explanation result that is consistent with the decision-making perspective of experts in the medical field.

[0196] The result interpretation device of the intelligent medical diagnosis system provided by the present application adopts the result interpretation method of the intelligent medical diagnosis system in the above-mentioned embodiment, and can solve the technical problem of result interpretation of the intelligent medical diagnosis system. Compared with the prior art, the beneficial effects of the result interpretation device of the intelligent medical diagnosis system provided by the present application are the same as the beneficial effects of the result interpretation method of the intelligent medical diagnosis system provided by the above-mentioned embodiment, and the other technical features in the result interpretation device of the intelligent medical diagnosis system are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0197] The present application provides a result interpretation device for an intelligent medical diagnostic system, the result interpretation device for the intelligent medical diagnostic system comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the result interpretation method for the intelligent medical diagnostic system in the above-mentioned embodiment one.

[0198] Reference below Figure 7 , which shows a schematic diagram of the structure of the result interpretation device of the intelligent medical diagnosis system suitable for implementing the embodiment of the present application. The result interpretation device of the intelligent medical diagnosis system in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The result interpretation device of the intelligent medical diagnosis system shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0199] like Figure 7As shown, the result interpretation device of the intelligent medical diagnosis system may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. Various programs and data required for the operation of the result interpretation device of the intelligent medical diagnosis system are also stored in RAM1004. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the result interpretation device of the intelligent medical diagnosis system to communicate wirelessly or wired with other devices to exchange data. Although the result interpretation device of the intelligent medical diagnosis system with various systems is shown in the figure, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0200] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0201] The result interpretation device of the intelligent medical diagnosis system provided by the present application adopts the result interpretation method in the above embodiment, which can solve the technical problem of result interpretation of the intelligent medical diagnosis system. Compared with the prior art, the beneficial effects of the result interpretation device of the intelligent medical diagnosis system provided by the present application are the same as the beneficial effects of the result interpretation method of the intelligent medical diagnosis system provided by the above embodiment, and the other technical features in the result interpretation device of the intelligent medical diagnosis system are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0202] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0203] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0204] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, the computer-readable program instructions being used to execute the result interpretation method of the intelligent medical diagnosis system in the above-mentioned embodiment.

[0205] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0206] The computer-readable storage medium may be included in the result interpretation device of the intelligent medical diagnosis system; or may exist independently without being assembled into the result interpretation device of the intelligent medical diagnosis system.

[0207] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the result interpretation device of the intelligent medical diagnosis system, the result interpretation device of the intelligent medical diagnosis system: obtains patient clinical data; inputs the patient clinical data into a pre-built cost-sensitive decision tree regularized neural network model; performs prediction processing on the patient clinical data through the neural network part in the cost-sensitive decision tree regularized neural network model to obtain a prediction result, and the cost-sensitive decision tree regularization part interprets the patient clinical data and the prediction result to obtain an interpretation result.

[0208] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0209] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0210] The modules involved in the embodiments described in the present application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0211] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the result interpretation method of the above-mentioned intelligent medical diagnosis system, and can solve the technical problem of result interpretation of the intelligent medical diagnosis system. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the result interpretation method of the intelligent medical diagnosis system provided in the above-mentioned embodiment, and will not be repeated here.

[0212] The present application also provides a computer program product, including a computer program, which implements the steps of the result interpretation method of the intelligent medical diagnosis system as described above when executed by a processor.

[0213] The computer program product provided by this application can solve the technical problem of interpreting the results of the intelligent medical diagnosis system. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the method for interpreting the results of the intelligent medical diagnosis system provided by the above embodiment, and will not be repeated here.

[0214] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for interpreting results of an intelligent medical diagnosis system, characterized in that: The method comprises: Obtain patient clinical data; Inputting the patient clinical data into a pre-built cost-sensitive decision tree regularized neural network model; The neural network part in the cost-sensitive decision tree regularized neural network model performs prediction processing on the patient's clinical data to obtain a prediction result, and the cost-sensitive decision tree regularization part performs interpretation processing on the patient's clinical data and the prediction result to obtain an interpretation result.

2. The method according to claim 1, characterized in that The step of inputting the patient clinical data into a pre-built cost-sensitive decision tree regularized neural network model also includes: Building a tree regularized neural network model based on a pre-collected training dataset; Constructing a cost-sensitive decision tree based on the training data set and the tree regularized neural network model; The cost-sensitive decision tree is added to the tree regularized neural network model to obtain a cost-sensitive decision tree regularized neural network model.

3. The method according to claim 2, characterized in that The step of constructing a tree regularized neural network model based on the pre-collected training data set includes: Obtain pre-collected training datasets and deep learning model frameworks; Inputting the training data set into the deep learning model framework to obtain a neural network prediction result; Constructing a binary classification decision tree according to the neural network prediction result and the training data set; A regularization term is calculated based on the binary classification decision tree and the deep learning model framework; The regularization term is added to the deep learning model for regularization processing to obtain a tree regularized neural network model.

4. The method according to claim 3, characterized in that After the step of adding the regularization term to the deep learning model for regularization processing to obtain a tree regularized neural network model, the method further includes: When the system has not reached the iteration upper limit, the training data is input into the tree regularized neural network model to obtain a new neural network prediction result, and the execution step is returned to: a binary classification decision tree is constructed according to the neural network prediction result and the training data set.

5. The method according to claim 3, characterized in that The step of constructing a cost-sensitive decision tree based on the training data set and the tree regularized neural network model comprises: Obtaining feature contribution based on the training data set and the tree regularized neural network model; A cost-sensitive decision tree is constructed according to the feature contribution.

6. The method according to claim 5, characterized in that The step of adding the cost-sensitive decision tree to the tree regularized neural network model to obtain the cost-sensitive decision tree regularized neural network model comprises: Obtaining a tree regularization term based on calculating the cost-sensitive decision tree; The tree regularization term is added to the tree regularized neural network model to obtain a cost-sensitive decision tree regularized neural network model.

7. The method according to claim 6, characterized in that After the step of adding the tree regularization term to the tree regularized neural network model to obtain the cost-sensitive decision tree regularized neural network model, the method further includes: When the system does not reach the iteration upper limit, return to the execution step: obtain the feature contribution based on the training data set and the tree regularized neural network model.

8. A result interpretation device for an intelligent medical diagnosis system, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the result interpretation method of the intelligent medical diagnosis system according to any one of claims 1 to 7.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the result interpretation method of the intelligent medical diagnosis system according to any one of claims 1 to 7 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the result interpretation method of the intelligent medical diagnosis system according to any one of claims 1 to 7 are implemented.

Images