Medical Data Processing Method and System

Abstract

The present disclosure relates to a medical data processing method and system. The method includes: obtaining medical data of a patient; inputting the medical data into a filling model to obtain corresponding predicted filling values; wherein the filling model is obtained by training a deep learning model based on sample medical data, and the sample medical data includes multiple different types of sample data. The training process includes: extracting context information of each sample data from the multiple different types of sample data, constructing a dependency graph structure based on the context information of each sample data, which includes multiple nodes and edges connecting adjacent nodes, each node represents a sample data and its context information, each edge represents the dependency relationship between the sample data of adjacent nodes, and the weight of the edge represents the similarity of the context information of the sample data of adjacent nodes; learning to output predicted filling values based on the dependency graph structure; filling the medical data with the predicted filling values corresponding to the missing values in the medical data to obtain new medical data.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the technical field of data processing, and in particular, to a medical data processing method and system. Background Art

[0002] Currently, the extensive application of high-dimensional data and multi-modal data of patients in the medical field provides rich information for clinical diagnosis and treatment. Based on artificial intelligence machine learning technology, it is possible to process the medical data information of patients to provide, for example, reference for diagnostic results, which requires training a machine learning model based on training data. The quality of the training data will affect the model performance, such as reducing the accuracy of the diagnostic results given by the model.

[0003] In related technologies, for the medical data used for model training, before model training, preprocessing such as missing value filling is usually performed. Since missing value problems are common in the medical data for training, traditional missing value filling methods often perform simple processing such as interpolation based on the overall distribution of a single data, which results in inaccurate filling results of the medical data, reduces the performance of the trained model, and thus reduces the trust and application ability of medical staff in the model in clinical practice. Summary of the Invention

[0004] To solve the above technical problems or at least partially solve the above technical problems, embodiments of the present disclosure provide a medical data processing method and system.

[0005] In a first aspect, embodiments of the present disclosure provide a medical data processing method, including:

[0006] Obtain the medical data of a patient;

[0007] Input the medical data into a filling model to obtain predicted filling values corresponding to the missing values in the medical data; wherein, the filling model is trained from a deep learning model based on sample medical data, the sample medical data includes multiple different types of sample data, and the training process of the deep learning model includes: extracting the context information of each sample data from the multiple different types of sample data, constructing a dependency graph structure based on the context information of each sample data, the dependency graph structure includes multiple nodes and edges connecting adjacent nodes; wherein, each node represents a sample data and its context information, each edge represents the dependency relationship between the sample data of adjacent nodes, and the weight of the edge represents the similarity of the context information of the sample data of adjacent nodes; learning to output predicted filling values based on the dependency graph structure until the convergence condition is met to end the training;

[0008] Fill the medical data with the predicted filling values corresponding to the missing values in the medical data to obtain new medical data.

[0009] In one embodiment, the training process of the deep learning model further includes:

[0010] Using a probabilistic graphical model to obtain the conditional probability distribution corresponding to each sample data in the dependency graph structure; optimizing and updating the dependency graph structure based on the conditional probability distribution corresponding to each sample data, and learning to output predicted filling values based on the updated dependency graph structure.

[0011] In one embodiment, the deep learning model includes a graph neural network, and the predicted filling values are learned and output based on the graph neural network for the dependency graph structure.

[0012] In one embodiment, the method further includes:

[0013] Obtaining new sample medical data and extracting the context information corresponding to the new sample medical data;

[0014] Updating the dependency graph structure based on the difference in similarity between the context information corresponding to the new sample medical data and the context information of each sample data;

[0015] Re-learning and outputting predicted filling values based on the updated dependency graph structure.

[0016] In one embodiment, the deep learning model adopts a self-supervised learning mechanism, and is provided with a first loss function and a second loss function. The first loss function defines that the goal of missing value filling is to minimize the loss value of the first loss function, so that the loss value of the first loss function is less than a set threshold; the second loss function is a contrastive loss function, and the contrastive loss function defines the difference in the feature similarity between the sample data before and after filling.

[0017] In one embodiment, the deep learning model adopts a Bayesian deep learning method, and the method further includes:

[0018] Calculating the uncertainty estimate value of the predicted filling value through Bayesian inference to assist medical personnel in evaluating the credibility of the filling result of medical data.

[0019] In one embodiment, the method further includes:

[0020] Obtaining the contribution value of each feature data in the medical data input into the filling model to the output predicted filling value through the model interpretation tool SHAP, so as to determine the most critical feature data in the decision-making process of the filling model;

[0021] And / or, introducing an attention mechanism into the filling model, so that the filling model uses the attention mechanism to automatically focus on and obtain the most important context information for the filling task.

[0022] In a second aspect, an embodiment of the present disclosure provides a medical data processing system, including:

[0023] An acquisition module, configured to acquire medical data of a patient;

[0024] A prediction module, configured to input the medical data into a filling model to obtain predicted filling values corresponding to missing values in the medical data; wherein, the filling model is obtained by training a deep learning model based on sample medical data, the sample medical data includes multiple different types of sample data, and the training process of the deep learning model includes: extracting context information of each sample data from the multiple different types of sample data, constructing a dependency graph structure based on the context information of each sample data, the dependency graph structure includes multiple nodes and edges connecting adjacent nodes; wherein, each node represents a sample data and its context information, each edge represents a dependency relationship between sample data of adjacent nodes, and the weight of the edge represents the similarity of the context information of the sample data of adjacent nodes; learning to output predicted filling values based on the dependency graph structure until the convergence condition is met to end the training;

[0025] A filling module, configured to fill the medical data with the predicted filling values corresponding to the missing values in the medical data to obtain new medical data.

[0026] In a third aspect, an embodiment of the present disclosure provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the medical data processing method described in any one of the above embodiments is implemented.

[0027] In a fourth aspect, an embodiment of the present disclosure provides an electronic device, including:

[0028] A processor; and

[0029] A memory, configured to store a computer program;

[0030] wherein, the processor is configured to execute the medical data processing method described in any one of the above embodiments by executing the computer program.

[0031] The technical solution provided by the embodiment of the present disclosure has the following advantages compared with the prior art:

[0032] The medical data processing method and system provided by the embodiments of the present disclosure obtain the medical data of a patient, and input the medical data into a filling model to obtain predicted filling values corresponding to the missing values in the medical data; wherein, the filling model is trained based on sample medical data for a deep learning model, the sample medical data includes multiple different types of sample data, and the training process of the deep learning model includes: extracting the context information of each sample data from the multiple different types of sample data, constructing a dependency graph structure based on the context information of each sample data, the dependency graph structure includes multiple nodes and edges connecting adjacent nodes; wherein, each node represents a sample data and its context information, each edge represents the dependency relationship between the sample data of adjacent nodes, and the weight of the edge represents the similarity of the context information of the sample data of adjacent nodes; learning to output predicted filling values based on the dependency graph structure until the convergence condition is met to end the training; filling the medical data with the predicted filling values corresponding to the missing values in the medical data to obtain new medical data. In this embodiment, the pre-trained filling model is used to accurately predict the filling values, which takes into account the local data characteristics of the data corresponding to the missing values in the medical data, such as the differences in context and the differences in the dependency relationships between different types of data, makes full use of the correlation between different types of data, reduces the information loss, and further improves the performance of the trained model, thereby improving the accuracy of the filling result of the final medical data. When training the model of this implementation scheme, the dependency graph structure is constructed by extracting the context information of the data to accurately capture the complex relationships between multi-source data, and then the precise filling of the missing values can be realized, which can effectively cope with the challenges of local feature differences and multi-modal data alignment, and utilize the accuracy and practicality of the results such as model disease diagnosis or efficacy evaluation based on the filled data in clinical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present disclosure and, together with the specification, are used to explain the principles of the present disclosure.

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0035] Figure 1 It is a flowchart of the medical data processing method according to the embodiments of the present disclosure;

[0036] Figure 2 It is a flowchart of the medical data processing method according to another embodiment of the present disclosure;

[0037] Figure 3 Schematic diagram of a medical data processing system according to another embodiment of the present disclosure;

[0038] Figure 4 Schematic diagram of an electronic device according to an embodiment of the present disclosure. Detailed implementation manners

[0039] In order to more clearly understand the above objects, features and advantages of the present disclosure, the solutions of the present disclosure will be further described below. It should be noted that, without conflict, the embodiments of the present disclosure and the features in the embodiments may be combined with each other.

[0040] In the following description, many specific details are set forth in order to fully understand the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only a part of the embodiments of the present disclosure, rather than all the embodiments.

[0041] It should be understood that in the following text, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" may mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B may be singular or plural. The character " / " generally means that the associated objects before and after are in an "or" relationship. "At least one (one) of the following" or its similar expression means any combination of these items, including any combination of single item (one) or plural items (ones). For example, at least one (one) of a, b or c may mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c may be single or multiple.

[0042] In the related art, multimodal data of patients, such as images of patients (such as CT images, X-ray images), physiological characteristic parameters, etc., have significant differences in terms of feature space, data format, time scale, etc., making effective multimodal data integration and alignment become increasingly complex. Therefore, the correlation between different modality data often cannot be fully utilized, resulting in information loss and a decrease in the accuracy of the data filling result. In addition, existing filling methods usually rely on the overall data distribution of a single type and cannot effectively cope with local feature differences and the alignment challenges of multimodal data, ignoring the complex dependence relationships between different types of data features, resulting in a decrease in the accuracy of the data filling result.

[0043] Figure 1 Flowchart of a medical data processing method according to an embodiment of the present disclosure. This method can be executed by a computing device such as a computer or a mobile terminal, and specifically may include the following steps:

[0044] Step S101: Obtain the medical data of the patient.

[0045] Exemplarily, the medical data of the patient can be high-dimensional data and multi-modal data, such as multi-modal (type) data like images, genomes, and electronic health records such as the patient's physiological characteristic data. Each type can include specific multiple data such as physiological characteristic data at different times, and there may be some missing values in the specific data corresponding to each type.

[0046] Step S102: Input the medical data into the filling model to obtain the predicted filling values corresponding to the missing values in the medical data. Among them, the filling model is trained based on sample medical data for a deep learning model. The sample medical data includes multiple different types of sample data. The training process of the deep learning model includes: extracting the context information of each sample data from the multiple different types of sample data, constructing a dependency graph structure based on the context information of each sample data. The dependency graph structure includes multiple nodes and edges connecting adjacent nodes; where each node represents a sample data and its context information, each edge represents the dependency relationship between the sample data of adjacent nodes, and the weight of the edge represents the similarity of the context information of the sample data of adjacent nodes; learning to output the predicted filling values based on the dependency graph structure until the convergence condition is met and the training ends.

[0047] Exemplarily, the sample medical data can include multiple different types of sample data such as the patient's physiological characteristic change data and disease texts such as the descriptive information in case records, etc. As the basis for constructing the dependency modeling, there is no restriction on the specific data type. Assume that the clinical dataset is X, where represents the sample data i (i is a natural number) of the m-th modality (or type), then its context information can be defined as It can be composed of multiple dimensional features, that is: where, c k represents the k-th context feature. For example, it can include the context features at the missing value in multiple sample data of the type to which the sample data i itself belongs, or it can include the context features associated with the missing value in other types of sample data.

[0048] Then construct a dependency graph structure G=(V, E). In this graph structure, the node V represents the sample data and its context information while the edge E represents the dependency relationship between different sample data, and the weight of the edge is determined by the similarity measure of the context information. As an example, the specific weight of the edge can be but is not limited to being determined by the following formula:

[0049]

[0050] Among them, is a Gaussian similarity function based on the context information of sample data i and j, and Σ represents the covariance matrix of the context information.

[0051] In this embodiment, during model training, by analyzing the context information in medical data, a dependency graph structure is constructed to accurately capture the interdependent relationships between different sample data. By calculating the similarity of the context information of sample data, an effective and accurate dependency graph structure is formed, providing preprocessed high-quality data for model training, improving the performance of the trained model, and thus enhancing the accuracy of the filling result of the final medical data. In this embodiment, for each type of data in medical data, the filling model can accurately predict missing values, which is beneficial for subsequent accurate filling. Among them, meeting the convergence condition can be that the loss value of the loss function of the deep learning model is less than a preset value.

[0052] Step S103: Fill the medical data with the predicted filling values corresponding to the missing values in the medical data to obtain new medical data.

[0053] This embodiment provides a context-driven missing value filling solution. During model training, by extracting the context information of each sample data and constructing a dependency graph structure, the integrity and accuracy of the data are ensured, significantly improving the performance and robustness of the filling algorithm model, and enabling the adaptive and accurate filling of missing values in medical data.

[0054] In this embodiment, an accurately pre-trained filling model is used to predict filling values. It takes into account the local data characteristics of the data types corresponding to the missing values in medical data, such as differences in context and differences in the dependency relationships between different data types, fully utilizes the correlation between different data types, reduces information loss, further improves the performance of the trained model, and thus enhances the accuracy of the filling result of the final medical data. During the training of this implementation scheme model, by extracting the context information of the data to construct a dependency graph structure to accurately capture the complex relationships between multi-source data, the precise filling of missing values is realized, which can effectively address the challenges of local feature differences and the alignment of multi-modal data, and utilize the accuracy and practicality of the results such as disease diagnosis or efficacy evaluation based on the filled data in clinical applications.

[0055] In one embodiment, as shown in combination with Figure 2 The training process of the deep learning model further includes steps S201 to S202: Step S201, using a probabilistic graphical model to obtain the conditional probability distribution corresponding to each sample data in the dependency graph structure; Step S202, optimizing and updating the dependency graph structure based on the conditional probability distribution corresponding to each sample data, and learning to output predicted filling values based on the updated dependency graph structure.

[0056] Exemplarily, a probabilistic graphical model is a theory that uses a graph to represent the probabilistic dependence relationships of variables. Combining the knowledge of probability theory and graph theory, a graph is used to represent the joint probability distribution of variables related to the model. In this embodiment, based on the dependence graph structure, a probabilistic graphical model is used to capture the dependence relationships between sample data, and the observed values of the sample data are set to have conditional dependence between their context information and the sample data of adjacent nodes, and the conditional probability distribution of the observed value can be expressed as:

[0057]

[0058] where neigh(i) represents the set of neighbor nodes of node i, is the sample data and its neighbor sample data is the distance function between them:

[0059]

[0060] where Λ is a positive definite matrix used to weight the distance, and the conditional probability distribution of the data similarity estimation weighted by the context information is obtained, thereby capturing the context dependence. The conditional probability distribution of is obtained, thereby capturing the context dependence.

[0061] In this embodiment, the probabilistic graphical model is applied to accurately obtain the conditional probability distribution corresponding to the sample data, and accordingly, the dependence graph structure is optimized and updated. That is, the previously established dependence graph structure may not be accurate enough, such as the weights of the edges are not precise enough. The dependence graph structure can be further optimized, such as the weights of the edges, through the conditional probability distribution, so that the dependence graph structure can more accurately capture the complex relationships between multi-source data, and further make the performance of the trained model better, realizing the accurate filling of missing values, and similar to optimizing the filling result to ensure the rationality and effectiveness of the filling strategy.

[0062] Based on any one of the above embodiments, in one embodiment, the deep learning model may include a Graph Neural Network, and based on the graph neural network GNN, the dependence graph structure is learned to output a predicted filling value.

[0063] Exemplarily, in this embodiment, the data of the dependence graph structure is constructed. Therefore, in order to further improve the performance of the trained model, an adaptive graph neural network is used for training, which can better extract and mine the data features of the dependence graph structure, so that the more accurate filling of missing values can be realized based on the finally trained filling model.

[0064] Based on any of the above embodiments, in one embodiment, the method may further include the following steps: obtaining new sample medical data, and extracting context information corresponding to the new sample medical data; updating the dependency graph structure based on the difference in similarity between the context information corresponding to the new sample medical data and the context information of each sample data; and re-learning and outputting predicted filled values based on the updated dependency graph structure.

[0065] Exemplarily, the context information corresponding to the new sample medical data may also be the context information of each sample data therein. The difference in similarity between the context information corresponding to the new sample medical data and the context information of each sample data in the sample medical data may be the difference in similarity of the context information of each pair of corresponding sample data in the two. When, for example, the differences corresponding to half of the sample data are greater than the set difference, the dependency graph structure may be updated, but this is not limited thereto.

[0066] In this embodiment, the dependency graph structure can be dynamically updated. After introducing the graph neural network, it can dynamically adjust the dependency graph structure for time changes and sample data updates. By measuring the similarity of the context features of new data samples, the dependency graph structure is updated in a timely manner, such as updating the node representation (i.e., the sample data and context information corresponding to the node change) and the weight of the edge, to ensure the adaptability of the model to medical data changes and the accuracy of prediction.

[0067] Specifically, in order to cope with the situation that relevant medical data, including the patient's condition, changes continuously over time, a graph neural network is introduced to achieve dynamic update of the graph structure. Let the graph structure at time t be G t =(V t , E t ). As time goes by, the updated graph structure is G t+1 =(V t+1 , E t+1 ). The update is based on the similarity measure between the context features of the new sample data and the context features of the existing sample data. By capturing the change in this similarity measure, such as the difference value of the similarity being greater than the preset similarity, the graph neural network can dynamically update the node representation and the edge weights.

[0068] In this way, this implementation scheme has the ability of dynamic update. Introducing the graph neural network enables the model to update the graph structure in real time and adapt to the changing medical data. For example, after the patient's condition changes, the previous model's prediction of the missing values in the patient's new medical data is no longer accurate. The model can be updated based on the new medical data as sample data. This feature ensures that the model can continuously adapt to the changes in the patient's condition, improving the diagnostic accuracy and analysis flexibility of the subsequent new medical data based on the filled values.

[0069] Based on any of the above embodiments, in one embodiment, the deep learning model may adopt a self-supervised learning mechanism and may be provided with a first loss function and a second loss function. The first loss function defines that the goal of missing value filling is to minimize the loss value of the first loss function, such that the loss value of the first loss function is less than a set threshold; the second loss function is a contrastive loss function, and the contrastive loss function defines the difference in the feature similarity between the sample data before and after filling.

[0070] Exemplarily, the goal of missing value filling is defined as minimizing the following first loss function:

[0071]

[0072] where, represents the true data of the i-th sample data in the m-th modality, is the filled value predicted by the model based on the context information , MSE represents the mean squared error, is the approximate posterior distribution of the latent variable , is the prior distribution of the latent variable, KL represents the Kullback-Leibler divergence, which is used to measure the difference between the approximate posterior distribution and the prior distribution, and λ is a hyperparameter that balances the MSE and the KL divergence.

[0073] In this embodiment, for improving the reliability of the filling strategy, a self-supervised learning mechanism is introduced to learn the filling strategy through the features of the sample data itself. The second loss function of the model, i.e., the contrastive loss function, can be set simultaneously. Based on the self-supervised learning method of contrastive loss, by comparing the sample data before and after filling, the feature similarity between similar samples is maximized, and the feature similarity between dissimilar samples is minimized, thereby improving the accuracy of the filling result. Exemplarily, the contrastive loss function can be defined by the following formula:

[0074]

[0075] where, 1(·) is an indicator function, and sim(H i , H j ) represents the feature similarity between the sample data i and j. The feature representation of the i-th sample data in the m-th modality is The feature representation of the overall sample data is H, where: During model training, the training can be ended when the loss value of the first loss function is less than a set threshold and the loss value of the second loss function is less than a set value.

[0076] In this embodiment, the filling target is defined as minimizing the first loss function to achieve efficient and accurate filling of missing values by the model trained based on constraints. Meanwhile, a self-supervised learning mechanism is introduced, and the model trained by the contrast loss function can optimize the filling strategy to improve the accuracy and reliability of the filling result.

[0077] In one embodiment, H obtained through joint embedding or contrast learning can also be used to ensure the consistency of the feature representations of sample data in different modalities in the shared space, thereby ensuring the effective fusion of sample data in different modalities and improving their feature consistency, and further improving the accuracy of the filling result.

[0078] Based on any of the above embodiments, in one embodiment, the deep learning model adopts the Bayesian deep learning method, which further includes: calculating the uncertainty estimate of the predicted filling value through Bayesian inference to assist medical personnel in evaluating the credibility of the filling result of medical data.

[0079] In this embodiment, by introducing the Bayesian deep learning method to estimate the uncertainty of the filling result, the reliability of the filling result can be further improved. Bayesian inference can provide uncertainty quantification for each filling value, thus helping medical personnel such as decision-makers evaluate the credibility of the filling result. Exemplarily, the uncertainty of the filling result can be defined as:

[0080] where Var[·] represents the variance of the latent variable and reflects the confidence interval of the filling result.

[0081] To further analyze the uncertainty, the total uncertainty can be decomposed into model uncertainty and data uncertainty. The inference error caused by insufficient training data or insufficient model complexity is approximately calculated by sampling the model parameters θ multiple times:

[0082]

[0083] The intrinsic noise or unmeasurable variability of the sample data itself is approximately calculated by the variance of the model directly for the observed data:

[0084]

[0085] The analysis through Bayesian deep learning provides the accuracy of the uncertainty estimate of the filling result, which enables medical professionals to enhance their trust in the model output results and helps to more widely apply this implementation in clinical practice.

[0086] Based on any of the above embodiments, in one embodiment, the method further includes: obtaining, by means of the model interpretation tool SHAP, the contribution value of each feature data in the medical data input to the filling model to the predicted filling value of the output, such as the SHAP value, so as to determine the most critical feature data in the decision-making process of the filling model, for example, the feature data with the largest SHAP value.

[0087] In this embodiment, the introduction of the SHAP value enhances the interpretability of the model, helps medical decision-makers evaluate the credibility of the filling result and the influence of the key feature data. Ensure that the filling result has good interpretability and practicability, thereby enhancing the trust of medical professionals in the data analysis results. Good interpretability can facilitate medical professionals to understand the process and results of data filling and integration, thereby enhancing the trust and application ability in clinical practice.

[0088] Based on any of the above embodiments, in one embodiment, an attention mechanism can be introduced into the filling model so that the filling model can automatically focus on obtaining the most important context information for the filling task by using this attention mechanism.

[0089] By introducing the attention mechanism, the model can automatically focus on the most important context information for the filling task. For example, through visualizing the attention weights, it can be intuitively understood how the model uses the context information of different sample data for filling, determine the most critical feature data, such as the feature data with the largest attention weight, improve the credibility and interpretability of the filling result, help medical decision-makers evaluate the credibility of the filling result and the influence of the key feature data. Ensure that the filling result has good interpretability and practicability, thereby enhancing the trust of medical professionals in the data analysis results, and thus solve the deficiencies of traditional methods in processing high-dimensional medical data and improve the quality and efficiency of clinical decision-making. Good interpretability can facilitate medical professionals to understand the process and results of data filling and integration, thereby enhancing the trust and application ability in clinical practice.

[0090] The solution in this embodiment is applicable to a complex high-dimensional data environment, can reduce and even avoid the problem of data incompleteness caused by data missing and the problem of reducing the performance of the machine learning model caused by inaccurate filling, and can improve the accuracy of the medical data filling result.

[0091] It should be noted that although the steps of the method in the present disclosure are described in a specific order in the accompanying drawings, this does not require or imply that these steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step for execution, and / or one step may be decomposed into multiple steps for execution, etc. Additionally, it is also easy to understand that these steps may be executed synchronously or asynchronously, for example, in multiple modules / processes / threads.

[0092] As Figure 3 shown, an embodiment of the present disclosure provides a medical data processing system, including:

[0093] An acquisition module 401, configured to acquire medical data of a patient;

[0094] A prediction module 402, configured to input the medical data into a filling model to obtain a predicted filling value corresponding to a missing value in the medical data; wherein, the filling model is obtained by training a deep learning model based on sample medical data, the sample medical data includes multiple different types of sample data, and the training process of the deep learning model includes: extracting context information of each sample data from the multiple different types of sample data, constructing a dependency graph structure based on the context information of each sample data, the dependency graph structure includes multiple nodes and edges connecting adjacent nodes; wherein, each node represents a sample data and its context information, each edge represents a dependency relationship between the sample data of adjacent nodes, and the weight of the edge represents the similarity of the context information of the sample data of adjacent nodes; learning to output a predicted filling value based on the dependency graph structure until the convergence condition is met to end the training;

[0095] A filling module 403, configured to fill the medical data with the predicted filling value corresponding to the missing value in the medical data to obtain new medical data.

[0096] In one embodiment, the training process of the deep learning model further includes: using a probabilistic graphical model to obtain the conditional probability distribution corresponding to each sample data in the dependency graph structure; optimizing and updating the dependency graph structure based on the conditional probability distribution corresponding to each sample data, and learning to output a predicted filling value based on the updated dependency graph structure.

[0097] In one embodiment, the deep learning model includes a graph neural network, and learning to output a predicted filling value based on the dependency graph structure by the graph neural network.

[0098] In one embodiment, the system further includes an update module for: obtaining new sample medical data and extracting context information corresponding to the new sample medical data; updating the dependency graph structure based on the difference in similarity between the context information corresponding to the new sample medical data and the context information of each sample data; and the deep learning model re-learning and outputting a predicted filling value based on the updated dependency graph structure.

[0099] In one embodiment, the deep learning model adopts a self-supervised learning mechanism and is provided with a first loss function and a second loss function. The first loss function defines that the goal of missing value filling is to minimize the loss value of the first loss function, such that the loss value of the first loss function is less than a set threshold; the second loss function is a contrastive loss function, and the contrastive loss function defines the difference in the feature similarity of the sample data before and after filling.

[0100] In one embodiment, the deep learning model adopts a Bayesian deep learning method, and the system may further include an estimation module for: calculating an uncertainty estimation value of the predicted filling value through Bayesian inference to assist medical personnel in evaluating the credibility of the filling result of the medical data.

[0101] In one embodiment, the system may further include an explanation module for: obtaining, through the model explanation tool SHAP, the contribution value of each feature data in the medical data input to the filling model to the output predicted filling value, so as to determine the most critical feature data in the decision-making process of the filling model.

[0102] In one embodiment, an attention mechanism is introduced into the filling model, so that the filling model can automatically focus on and obtain the context information most important for the filling task by using this attention mechanism.

[0103] Regarding the system in the above embodiments, the specific manners in which each module performs operations and the corresponding technical effects brought have been described in detail correspondingly in the embodiments related to the method, and will not be elaborated here.

[0104] It should be noted that although several modules or units of a device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present disclosure, the features and functions of two or more of the above-described modules or units can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units. A component shown as a module or unit may or may not be a physical unit, that is, it may be located in one place, or may be distributed over multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present disclosure. A person of ordinary skill in the art can understand and implement it without creative work.

[0105] An embodiment of the present disclosure also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the medical data processing method of any one of the above embodiments is implemented.

[0106] Exemplarily, the readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection with one or more wires, a portable 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.

[0107] The computer-readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, in which the readable program code is carried. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium can also be any readable medium other than the computer-readable storage medium, and the readable medium can send, propagate, or transmit a program used by or in combination with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted by any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0108] An embodiment of the present disclosure also provides an electronic device, including a processor and a memory, where the memory is used to store a computer program. Wherein, the processor is configured to execute the medical data processing method in any one of the above embodiments by executing the computer program.

[0109] Next, refer to Figure 4Describe the electronic device 600 according to this embodiment of the present invention. Figure 4 The displayed electronic device 600 is merely an example and should not impose any limitations on the functions and usage scope of the embodiments of the present invention.

[0110] As Figure 4 shown, the electronic device 600 is presented in the form of a general-purpose computing device. The components of the electronic device 600 may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different system components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0111] Among them, the storage unit stores program code, and the program code can be executed by the processing unit 610, so that the processing unit 610 executes the steps according to various exemplary embodiments of the present invention described in the method embodiment part of this specification above. For example, the processing unit 610 can execute the steps of the method as Figure 1 shown.

[0112] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 6201 and / or a cache storage unit 6202, and may further include a read-only storage unit (ROM) 6203.

[0113] The storage unit 620 may also include a program / utilities 6204 having a set (at least one) of program modules 6205. Such program modules 6205 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. The implementation of a network environment may be included in each or some combination of these examples.

[0114] The bus 630 may represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any bus structure in a variety of bus structures.

[0115] The electronic device 600 can also communicate with one or more external devices 700 (such as a keyboard, a pointing device, a Bluetooth device, etc.), and can also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or communicate with any device (such as a router, a modem, etc.) that enables the electronic device 600 to communicate with one or more other computing devices. Such communication can be carried out through the input / output (I / O) interface 650. Moreover, the electronic device 600 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through the network adapter 660. The network adapter 660 can communicate with other modules of the electronic device 600 through the bus 630. It should be understood that although not shown in the figure, other hardware and / or software modules can be used in combination with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0116] Through the description of the above embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software, or can be implemented by the way of software combined with necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product, and the software product can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, or a network device, etc.) to execute the method steps of the above various embodiments according to the embodiments of the present disclosure.

[0117] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without more limitations, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0118] The above are only specific embodiments of the present disclosure, enabling those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments described herein, but rather will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A medical data processing method, characterized in that: include: Access to patients’ medical data; The medical data is input into a filling model to obtain a predicted filling value corresponding to the missing value in the medical data; wherein the filling model is obtained by training a deep learning model based on sample medical data, and the sample medical data includes multiple different types of sample data, and the training process of the deep learning model includes: extracting context information of each sample data from the multiple different types of sample data, and constructing a dependency graph structure based on the context information of each sample data, wherein the dependency graph structure includes multiple nodes and edges connecting each adjacent node; wherein each node represents a sample data and its context information, each edge represents a dependency relationship between sample data of adjacent nodes, and the weight of the edge represents the similarity of the context information of sample data of adjacent nodes; learning and outputting predicted filling values ​​based on the dependency graph structure until the convergence condition is met and the training is terminated; Based on the predicted filling values ​​corresponding to the missing values ​​in the medical data, the medical data are filled to obtain new medical data.

2. The method according to claim 1, characterized in that: The training process of the deep learning model also includes: A probability graph model is used to obtain the conditional probability distribution corresponding to each sample data in the dependency graph structure; the dependency graph structure is optimized and updated based on the conditional probability distribution corresponding to each sample data, and a predicted fill value is output based on the learning of the updated dependency graph structure.

3. The method according to claim 1, characterized in that The deep learning model includes a graph neural network, which learns and outputs a predicted fill value based on the dependency graph structure.

4. The method according to claim 3, characterized in that: The method further includes: Acquire new sample medical data, and extract context information corresponding to the new sample medical data; Update the dependency graph structure based on the difference in similarity between the context information corresponding to the new sample medical data and the context information of each sample data; The predicted fill value is re-learned based on the updated dependency graph structure.

5. The method according to claim 1, characterized in that The deep learning model adopts a self-supervised learning mechanism and is provided with a first loss function and a second loss function. The first loss function defines that the goal of missing value filling is to minimize the loss value of the first loss function so that the loss value of the first loss function is less than a set threshold; the second loss function is a contrast loss function, which defines the difference in feature similarity of sample data before and after filling.

6. The method according to claim 1, characterized in that The deep learning model adopts a Bayesian deep learning method, and the method also includes: The uncertainty estimate of the predicted filling value is calculated through Bayesian reasoning to assist medical personnel in evaluating the credibility of the filling results of medical data.

7. The method according to any one of claims 1 to 6, characterized in that: The method further includes: The contribution value of each feature data in the medical data input into the filling model to the output predicted filling value is obtained through the model interpretation tool SHAP, so as to determine the most critical feature data of the filling model in the decision-making process; And / or, introducing an attention mechanism into the filling model so that the filling model uses the attention mechanism to automatically focus on obtaining the most important contextual information for the filling task.

8. A medical data processing system, characterized in that: include: An acquisition module, used to obtain the patient's medical data; A prediction module, used for inputting the medical data into a filling model to obtain a predicted filling value corresponding to the missing value in the medical data; wherein the filling model is obtained by training a deep learning model based on sample medical data, and the sample medical data includes a plurality of different types of sample data, and the training process of the deep learning model includes: extracting context information of each sample data from the plurality of different types of sample data, and constructing a dependency graph structure based on the context information of each sample data, wherein the dependency graph structure includes a plurality of nodes and edges connecting adjacent nodes; wherein each node represents a sample data and its context information, each edge represents a dependency relationship between sample data of adjacent nodes, and the weight of the edge represents the similarity of the context information of sample data of adjacent nodes; learning and outputting a predicted filling value based on the dependency graph structure until the convergence condition is met and the training is terminated; A filling module is used to fill the medical data to obtain new medical data based on the predicted filling value corresponding to the missing value in the medical data.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the medical data processing method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: processor; as well as Memory for storing computer programs; Wherein, the processor is configured to perform the medical data processing method according to any one of claims 1 to 7 by executing the computer program.

Images