Medical image data processing method, system and electronic device for realizing data security

Abstract

The application provides a medical image data processing method and system for realizing data security and an electronic device, and relates to the technical field of medical images. The method comprises the following steps: performing data checking and anonymization processing on original data to generate to-be-processed data; predicting by using a decision model according to user data processing requirements to generate a data analysis execution flow and an interaction frequency parameter; the data analysis execution flow comprises a data processing link and a modeling link; processing the to-be-processed data by an execution server to complete data analysis of the data processing link; and constructing a corresponding model based on the data analysis execution flow to generate a server comprehensive model, which is used for predicting input data and generating a final processing result in combination with the data processing link, so as to realize automatic and safe analysis and processing of medical image data, alleviate the technical problems of low data security and low processing efficiency in the prior art, and achieve the technical effects of improving data processing security and processing efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical imaging, in particular to a medical image data processing method and system for realizing data security and an electronic device. BACKGROUND

[0002] Medical image data (such as neural image data and other medical data) is usually obtained from hospitals. When training an AI model using medical image data from a hospital, the trained model usually has insufficient generalization ability due to the regional distribution characteristics of the population covered by each hospital, which cannot be better applied to medical practice. To solve this problem, a general high-performance model is usually trained by aggregating medical image data from different hospitals. However, due to ethical and related rules (safety considerations), hospitals cannot share relevant medical image data. At the same time, due to differences in equipment, work habits, and other factors, medical image data from each hospital also faces different characteristics.

[0003] In addition, in the data processing process of medical images (especially neural images), different processing steps such as preprocessing, feature extraction, statistical analysis, modeling, etc. are usually required to combine different research goals. These processing steps will involve numerous software packages. In order to correctly and efficiently use these software, data processing personnel need to deeply understand the algorithms, parameters, result formats and meanings, design methods, etc. behind these software. For researchers in neuroscience, the threshold for use is high, the learning difficulty is great, and the time-consuming is long. That is, the existing data processing scheme combining medical image data from multiple hospitals has the problems of low data security, high processing difficulty, and low processing efficiency. SUMMARY

[0004] The purpose of the present application is to provide a medical image data processing method and system for realizing data security and an electronic device to alleviate the technical problems of low data security, high processing difficulty, and low processing efficiency in the prior art.

[0005] In a first aspect, an embodiment of the present application provides a medical image data processing method for realizing data security, applied to a server, the server comprising an execution server and a core server, and the method comprising: performing data checking and anonymization processing on pre-acquired original data to generate to-be-processed data;

[0006] According to the user data processing requirement, a pre-generated decision model is used for prediction to generate a data analysis execution flow and an interaction frequency parameter; wherein the decision model includes a data analysis decision model and an interaction frequency parameter automatic recommendation decision model; the interaction frequency parameter is used to represent the interaction frequency between the execution server and the core server; the data analysis execution flow includes a plurality of medical image data analysis links and a corresponding sequence; the medical image data analysis link includes a data processing link and a modeling link;

[0007] Based on the data analysis execution flow, the execution server processes the to-be-processed data to complete the data analysis of the data processing link;

[0008] Based on the data analysis execution flow, a server comprehensive model is generated to complete the model construction of the modeling link; the server comprehensive model is used to predict the input data and generate a final processing result in combination with the data processing link to realize automatic and safe analysis and processing of medical image data.

[0009] In some possible implementation manners, the step of generating to-be-processed data by performing data checking and anonymization processing on the pre-acquired original data includes: checking the pre-acquired original data according to a predefined data format; if the original data does not conform to the predefined data format, performing format conversion on the original data according to the predefined data format; checking the data attribute of the original data to determine that the original data meets the analysis condition; processing the original data under different acquisition parameters to convert the original data into standard data; performing de-privatization processing on the header information and facial feature information of the standard data to generate to-be-processed data.

[0010] In some possible implementation manners, the user data processing requirement includes a data processing purpose, original data, data processing expectation, storage and servers participating in calculation; before the steps of according to the user data processing requirement, using a pre-generated decision model for prediction to generate a data analysis execution flow and an interaction frequency parameter, the method further includes: generating a training database based on a pre-constructed medical image data processing knowledge base; the framework of the medical image data processing knowledge base includes a plurality of first-level categories; the first-level categories include a processing target, object data, an analysis link, a file, and a task; using the training database to train the data analysis decision model to generate a trained data analysis decision model.

[0011] In some possible implementation manners, before the steps of generating the data analysis execution flow and the interaction frequency parameter according to the user data processing requirement and using the pre-generated decision model for prediction, the method further includes: obtaining a pre-defined rule between the interaction frequency parameter, the user data processing requirement, the number of servers participating in calculation, and the data analysis execution flow; the pre-defined rule is used to recommend the interaction frequency parameter as a basis for relevant decisions; constructing a decision recommendation model based on relevant data in a server running process by using a machine learning method; input of the decision recommendation model is the encoded user data processing requirement, the number of servers, and the data analysis execution flow; output of the decision recommendation model is the interaction frequency parameter; and determining an interaction frequency parameter automatic recommendation decision model based on the pre-defined rule and the decision recommendation model.

[0012] In some possible implementation manners, the steps of generating the data analysis execution flow and the interaction frequency parameter according to the user data processing requirement and using the pre-generated decision model for prediction include: generating the data analysis execution flow by using a trained data analysis decision model for prediction according to the user data processing requirement; and generating the interaction frequency parameter by using the interaction frequency parameter automatic recommendation decision model for prediction.

[0013] In some possible implementation manners, the step of generating the server comprehensive model based on the data analysis execution flow includes: each execution server performs corresponding model training for the modeling link, and if the training process meets the interaction frequency parameter requirement, the current model parameter is sent to the core server; the core server generates comprehensive model parameters based on the received current model parameters, and sends the comprehensive model parameters to each execution server; each execution server performs training based on the comprehensive model parameters, and the parameter interaction process is repeatedly performed until the current model meets a corresponding standard, and the current model is determined as the server comprehensive model.

[0014] In some possible implementation manners, the method further includes: calculating a data quality control index for a processing result of each link in the data analysis execution flow; the data quality control index is used to evaluate the usability of data; and generating a result evaluation parameter for the final processing result.

[0015] In a second aspect, an embodiment of the present application provides a medical image data processing system for realizing data security, applied to a server, the server including: an execution server and a core server; and the system including: a data generation module, configured to perform data checking and anonymization processing on pre-acquired original data to generate to-be-processed data;

[0016] A model prediction module is configured to generate a data analysis execution flow and an interaction frequency parameter according to a user data processing requirement and by using a pre-generated decision model, wherein the decision model comprises a data analysis decision model and an interaction frequency parameter automatic recommendation decision model, the interaction frequency parameter is used to represent an interaction frequency between the execution server and the core server, and the data analysis execution flow comprises a plurality of medical image data analysis links and a corresponding sequence, and the medical image data analysis link comprises a data processing link and a modeling link;

[0017] A data processing module is configured to process the to-be-processed data by the execution server based on the data analysis execution flow, so as to complete data analysis of the data processing link.

[0018] A result generation module is configured to generate a server comprehensive model by constructing a corresponding model based on the data analysis execution flow, so as to complete model construction of the modeling link, the server comprehensive model is used to predict input data and generate a final processing result in combination with the data processing link, so as to realize automatic and safe analysis and processing of medical image data.

[0019] In a third aspect, an electronic device is provided, which comprises a memory and a processor, the memory stores a computer program capable of running on the processor, and the processor implements the steps of the method of any one of the first aspect when executing the computer program.

[0020] In a fourth aspect, a computer readable storage medium is provided, which stores computer executable instructions, and the computer executable instructions, when invoked and run by a processor, cause the processor to run the method of any one of the first aspect.

[0021] The application provides a medical image data processing method, system and electronic equipment for realizing data security. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the specific embodiments or prior art of the present application, the drawings needed to be used in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0023] Figure 1 A flowchart of a medical image data processing method for realizing data security provided by an embodiment of the present application;

[0024] Figure 2 A flowchart of a medical image data processing method for realizing data security provided by an embodiment of the present application;

[0025] Figure 3 A structural diagram of a medical image data processing system for realizing data security provided by an embodiment of the present application;

[0026] Figure 4 A structural diagram of an electronic equipment provided by an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0028] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0029] It should be noted that similar reference numbers and letters represent similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Some embodiments of the present application are described in detail below with reference to the accompanying drawings. The following embodiments and features in the embodiments can be combined with each other without conflict.

[0030] Medical image data (such as neural image data and other medical data) is usually obtained from hospitals. When training an AI model using medical image data from a hospital, the trained model usually has insufficient generalization ability due to the regional distribution characteristics of the population covered by each hospital, and therefore cannot be better applied to medical practice. To solve this problem, the general solution is to aggregate medical image data from different hospitals to jointly train a high-performance comprehensive model. However, due to ethical and related rules (safety considerations), hospitals cannot share relevant medical image data; at the same time, due to differences in equipment, work habits, etc. of each hospital, the medical image data of each hospital also faces the problem of different characteristics.

[0031] In addition, in the data processing process of medical images (especially neural images), different research targets are usually combined to perform processing and calculation of different links, such as preprocessing, feature extraction, statistical analysis, modeling, etc., and these processing links will involve numerous software packages. According to incomplete statistics, there are at least hundreds of neural image related software. In order to correctly and efficiently use these software, data processing personnel need to deeply understand the algorithms, parameters, result formats and meanings, design methods, etc. behind these software. For researchers in neuroscience, since they often have less software programming ability, this should be a very difficult problem, especially when different software needs to be integrated to form a processing flow, and the matching between software needs to be deeply understood, which are difficult and time-consuming problems. That is, the existing scheme of combining medical image data of each hospital for data processing has the problems of low data security, high processing difficulty and low processing efficiency.

[0032] Based on this, the embodiment of the present application provides a medical image data processing method, system and electronic equipment for realizing data security to alleviate the above problems.

[0033] In order to facilitate the understanding of the present embodiment, first of all, a medical image data processing method for realizing data security disclosed by the present embodiment is introduced in detail, referring to the flow diagram of a medical image data processing method for realizing data security shown in Figure 1 The method is applied to a server, and the server includes a client server and a server server. The method can be executed by an electronic device, and mainly includes the following steps S110 to S140:

[0034] S110: performing data checking and anonymization processing on the pre-acquired original data to generate to-be-processed data;

[0035] In order to effectively realize the medical image data processing for realizing data security, considering that the data of each data owner may be different, the data needs to be processed first. Mainly including data format conversion, data selection, data conversion, data anonymization, etc.

[0036] In the present embodiment, the step S110 specifically includes:

[0037] (1) First, the pre-acquired original data is checked according to the predefined data format; if the original data does not conform to the predefined data format, the original data is converted according to the predefined data format; that is, data format conversion: for the data object in the storage, check according to the pre-defined data organization method (such as bids), if it conforms to the agreed format, pass the format check; if the data does not conform to the agreed format, the data object is converted according to the agreed format;

[0038] (2) Check the data properties of the original data to determine whether the original data meets the analysis conditions; that is, data selection: mainly check whether the participating neural image data meets the required modalities, quality requirements, etc.

[0039] (3) Process the original data under different acquisition parameters to convert the original data into standard data; that is, data conversion: for different situations existing in the participating neural image data, such as different field strengths, different resolutions, etc., use adversarial generative networks, variational autoencoders and flow models to convert the data into a relatively unified condition (while preserving the original data);

[0040] (4) Perform de-privatization processing on the header information and facial feature information of the standard data to generate the data to be processed. That is, data anonymization processing: on the one hand, the header information of the data is processed to avoid the existence of names and other private information, and on the other hand, the facial feature removal method based on the face mask is used to ensure the anonymization of the facial features; thus, the data checking work is completed.

[0041] The unified data format, data quality, etc. after the data checking here lay a data foundation for subsequent flexible selection, automatic execution of data processing flow, and efficient data processing modeling. The above data checking work is automatically executed by the unified script located on each execution server.

[0042] S120: According to the user data processing requirements, the decision model is used to predict the data analysis execution flow and the interaction frequency parameter.

[0043] In this embodiment, the user data processing requirements include: data processing purpose, original data, data processing expectation, participating computing storage and server.

[0044] To maximize the convenience of users to process medical image related data, the method provides a graphical interface for users to express medical image data processing related requirements. The content of the requirements expressed by the user in the graphical interface can be divided into four categories: data processing purpose, data itself (i.e. original data), data processing expectation, storage and server participating in calculation, etc. (for security considerations, the servers and storages that can be calculated are specified in advance, and the data has been saved in the servers and storages); referring to the knowledge base framework described below, the data processing purpose content mainly includes: the disease targeted (which can be empty), the focus, etc. The data itself related content includes: data size, format, mode, current location (as well as directory hierarchy relationship - such as subject, group, etc.), attribute (training, testing, verification), etc. The data processing expectation related content includes: expected data processing time, space to be occupied, expected performance indicators, etc. The storage and server content participating in calculation includes storage name, type, location, capacity, ip, and execution server name, type, location, ip, etc. (overall, the servers performing secure calculation are divided into two categories: core servers (server) and execution servers (client), the former is unique, and the latter is multiple, and the training, verification, and testing data sets required for modeling are also specified here).

[0045] In the embodiment, the decision model includes: a data analysis decision model and an interactive frequency parameter automatic recommendation decision model; the interactive frequency parameter is used to represent the interaction frequency between the execution server and the core server; the data analysis execution flow includes a plurality of medical image data analysis links and a corresponding order; the medical image data analysis link includes: a data processing link and a modeling link.

[0046] In one embodiment, the above method further includes: generating a training database based on a pre-constructed medical image data processing knowledge base; the framework of the medical image data processing knowledge base includes a plurality of first-level categories; the first-level categories include: processing target, object data, analysis link, file, task; using the training database to train the data analysis decision model to generate a trained data analysis decision model.

[0047] To realize data security medical image data processing calculation and training database construction, first construct a medical image data processing knowledge base, and its framework is constructed using a multi-level multi-link method. As a specific example, the category relationship of the knowledge base framework is as follows: the first-level categories include: processing target, object data, calculation process (i.e. analysis link), file, task, etc.

[0048] The secondary categories are subdivided from the primary categories, such as: the categories under processing target include: disease, concern, efficiency, performance, etc., the categories under object data include: original, conversion, intermediate result, final result, etc., the categories under calculation process include: preprocessing, feature extraction, statistical analysis, machine learning (shallow) modeling, deep learning modeling, visualization, data quality control, result annotation, auxiliary tool, processing strategy, etc., the categories under file include: configuration file, data file, log file, etc., the categories under task include: file transfer task, data processing task, etc.

[0049] The third-level categories are subdivided from the second-level categories, for example, the categories under disease can include dementia, autism, etc., the categories under focus can include cognition, social cognition, emotion, quality of life, body function, etc., the categories under efficiency can include operation time, space occupation, etc., the categories under performance can include accuracy, recall rate, significance level, etc., the categories under raw data include demographic characteristics, group, mode, field strength, sequence, additional features, related data, etc., the categories under conversion / intermediate results include steps, inputs, algorithms, results, etc., the categories under final results include statistical analysis results (including subjects / groups, algorithms, results), model modeling results (including subjects / groups, algorithms, models), etc., the categories under preprocessing include format conversion, removal of time points, time layer correction, head motion correction, standardization, regression covariates, linear drift removal, smoothing, filtering, deletion of time points, generation, etc., the categories under feature extraction include statistical indicators, graph theory indicators, structural networks, functional networks, feature selection and dimensionality reduction, etc., the categories under statistical analysis can include difference, correlation, regression, etc., the categories under machine learning modeling can include classification, clustering, integration, etc., the categories under deep learning can include cnn-based models, rnn-based models, attention-based models, integrated models, etc., the categories under data quality control can include quality control indicators, quality control methods, etc., the categories under result annotation can include annotation methods, annotation algorithms, etc., the categories under auxiliary tools can include format conversion, anonymization, etc., the categories under processing strategy can include analysis strategy, calculation strategy, etc., the categories under data files can include raw files, conversion files, intermediate process files, calculation result files, etc., the categories under data processing tasks can include data conversion tasks, preprocessing tasks, feature extraction tasks, statistical analysis tasks, machine learning tasks, deep learning tasks, visualization tasks, etc. The fourth-level, fifth-level, etc. categories are obtained by further subdividing in turn, for example, the cognition category can be further subdivided into calculation, reasoning, problem solving, decision making, perception, memory, attention, visual space, execution, learning, language, the social cognition category can be further subdivided into interpersonal relationship, social adaptation, etc., the emotion category can be further subdivided into anxiety, depression, hostility, retardation, confusion, etc., the graph theory indicator category can be further subdivided into node degree, shortest path, clustering coefficient, global efficiency, rich hub, etc., the structural network category can be further subdivided into white matter fiber, structural covariate, structural causality, etc., the functional network category can be further subdivided into dynamic, resting state, and undirected, directed, etc., the quality control indicator category can be further subdivided into signal-to-noise ratio, difference between structural images and templates, brain tissue symmetry, gray matter outside the template, difference between tool results, etc., the quality control method category can be further subdivided into supervised method, unsupervised clustering, etc., the analysis strategy category can be further subdivided into separate analysis, voting analysis, multi-dimensional analysis, etc., and so on until it cannot be subdivided.

[0050] In addition, each subcategory under the calculation process category contains input, output, parameters, algorithms, tools, steps, templates (if any), and other detailed parameters; each subcategory under the object data category includes data total capacity, format, location, small file proportion, and other subcategories.

[0051] The relationship categories in the knowledge base framework mainly include is-a and attribute relationships. According to the above knowledge base framework, using the evidence-based literature approach, for the processing target, object data, calculation process, and other content records that have clear and high-quality evidence support, a knowledge base is established (the entity and relationship knowledge content can be obtained by information extraction method or artificial sorting method, and the constructed knowledge base is essentially a knowledge graph).

[0052] In this embodiment, to obtain the data analysis decision model, first, a training database is constructed, and the training data comes from the content in the above medical image data processing knowledge base and is confirmed by expert annotation. The specific data mainly includes analysis requirements (mainly including disease, focus point-constituting an encoded text sequence after standardization, and calculation efficiency and performance requirements-corresponding to extremely high, high, medium, etc. options according to the pre-determined rules-and constituting an encoded text sequence) and data analysis execution flow; the data analysis execution flow is composed of one or more components, and the component mainly refers to preprocessing, feature extraction, statistical analysis, machine learning modeling, deep learning modeling, visualization, result annotation, processing strategy, etc.; the component is composed of one or more modules, and the module refers to a specific algorithm applied to a specific calculation, such as head motion correction algorithm (which can be multiple) in preprocessing; from this perspective, the data analysis execution flow is a directed graph, the node is the module, and the edge is the sequential relationship; the text sequence of the data analysis requirement and

[0053] The text sequence of the data analysis execution flow (all after standardization) is represented in the following way; and the training data is divided into training, validation, and test sets in the ratio of 7:1.5:1.5.

[0054] The encoded text sequence expression of the analysis requirement is composed of: the disease, focus point, etc. after standardization constitute an encoded text sequence, the calculation efficiency, performance requirement, etc. are corresponding to extremely high, high, medium, etc. options according to the pre-determined rules and constitute an encoded text sequence, and then

[0055] The text sequences are combined together to form a unified text sequence (the connection symbol is a comma,,), such as: (Alzheimer's disease, memory, high efficiency, high performance). The data analysis execution flow text sequence expression is composed of

[0056] wherein the main part is component, component includes module, specific algorithm and related parameter setting recommendation (connected by comma), adopts hierarchical method to show, uses “->” between components to show sequential relationship, uses “= >” between modules to show sequential relationship, and uses comma “,” to show parallel relationship, such as:

[0058] 5(preprocessing

[0060] (remove time point, algorithm =, first = 10,…) = > (…)

[0063] 0->(feature extraction (…) = >…)

[0064] ->(…)

[0066] Correspondingly, the input of the data analysis decision model is the related requirements of data analysis, that is, the literal sequence expression composed of disease, focus, calculation efficiency and performance requirements (after standardization). The output of the data analysis decision model is the literal sequence expression composed of components (i.e. preprocessing, feature extraction, statistical analysis, machine learning modeling, deep learning modeling, visualization, result annotation, processing strategy and specific modules constituting the components). The model adopts a sequence-to-sequence generation model, such as rnn combined with attention or T5, etc.

[0067] According to the data analysis requirements of the user, the data analysis execution flow can be obtained from the data analysis decision model. The software related preset parameters in the data analysis execution flow are preset according to the evidence-based literature, and the source is marked.

[0068] The data object related preset parameters are obtained by pre-detecting the software (including the user preset part). In addition, the interaction frequency between each execution server and the core server server participating in the calculation is a very important parameter, which will affect the effect and time-consuming of data analysis modeling, etc., so the parameter is also automatically recommended and decided.

[0069] In one embodiment, the above method further comprises: obtaining the interaction frequency parameter and the user data processing requirement, the number of servers participating in calculation and the pre-defined rule between the data analysis execution flow; the predefined rule is used as the basis for related decision to recommend the interaction frequency parameter; based on the running process of the server

[0070] ​​​​​The decision recommendation model is constructed by using a machine learning method based on the related data in the running process of the server. The input of the decision recommendation model is the encoded user data processing requirement, the number of servers, and the data analysis execution flow. The output of the decision recommendation model is the interaction frequency parameter. The interaction frequency parameter automatic recommendation decision model is determined based on the pre-defined rule and the decision recommendation model.

[0071] That is, the interaction frequency parameter automatic recommendation decision model includes two parts: a rule part and a machine learning model part. First, the rules between the interaction frequency parameter and the related requirements of user processing data, the number of servers, and the data analysis execution flow determined by the recommendation are established according to the related expert knowledge, and the related decisions are made according to the rules. Based on the related data in the running process of the server, a related model between the interaction frequency parameter and the related requirements of user processing data, the number of servers, and the data analysis execution flow determined by the recommendation is constructed by using a machine learning method (the training data comes from the running data confirmed by expert annotation), and the model effect reaches the related requirements. The model executes the decision recommendation according to the model after the model effect reaches the related requirements. The algorithm used includes neural network-based integration. The input of the model is the related requirements of user processing data, the number of servers, and the data analysis execution flow determined by the recommendation (the related parameters are encoded), and the output of the model is the interaction frequency. The loss function of the model considers the minimum execution time and the optimal model effect as the target. The determined result is expressed in a graphical manner, and a self-customization selection is provided.

[0072] In the embodiment, the step of S120 specifically includes: generating a data analysis execution flow by predicting, using a trained data analysis decision model, according to the user data processing requirement; and generating an interaction frequency parameter by predicting, using an interaction frequency parameter automatic recommendation decision model.

[0073] That is, according to the related requirements of user processing data in the information configuration, including diseases, concerns, and data conditions, based on the established data analysis decision model, the specific execution procedure of the data analysis flow is obtained, including preprocessing, feature extraction, statistical analysis, machine learning modeling, deep learning modeling, result annotation, processing strategy, and the like, and the parameters such as the interaction frequency between the execution server (client) and the core server (server) generated based on the interaction frequency parameter automatic recommendation decision model.

[0074] After the user customizes and confirms the data analysis execution flow, the data analysis execution flow (not including machine learning modeling and deep learning modeling) is executed on each execution server, and the data analysis process is automatically completed. Then each execution server starts to execute the system modeling process (machine learning modeling and deep learning modeling), and periodically interacts with the core server for model parameters until the model reaches the corresponding standard, and the modeling process is stopped.

[0075] S130: executing the data analysis execution flow to process the to-be-processed data by the execution server to complete data analysis of the data processing link;

[0076] That is, the local execution process, after the user customizes and confirms the data analysis execution flow, the data analysis execution flow is executed on each execution server (without machine learning modeling and deep learning modeling), and the data analysis process (such as general data preprocessing, feature extraction, and statistical analysis) is automatically completed.

[0077] The intermediate result files and result files in the data analysis process are stored in a predetermined format (such as referring to bids) (annotated with date, software name, etc. for distinction).

[0078] S140: based on the data analysis execution flow, corresponding model construction is performed to generate a server comprehensive model to complete model construction of the modeling link; the server comprehensive model is used for predicting input data, and the final processing result is generated in combination with the data processing link to realize automatic and safe analysis and processing of medical image data.

[0079] In one embodiment, the step of generating a server comprehensive model based on the data analysis execution flow in S140 includes: each execution server performs corresponding model training for the modeling link, and if the training process meets the interaction frequency parameter requirement, the current model parameter is sent to the core server; the core server generates comprehensive model parameters based on the received current model parameters, and sends the comprehensive model parameters to each execution server; each execution server performs training based on the comprehensive model parameters, and the parameter interaction process is repeated until the current model meets the corresponding standard, and the current model is determined as the server comprehensive model.

[0080] That is, the collaborative modeling process, according to the data analysis execution flow confirmed by the user customization, each execution server starts to execute the system modeling process (machine learning modeling and deep learning modeling): according to the division of the training, verification, and test data sets specified in the information configuration, the machine learning (or deep learning) algorithm determined in the data analysis execution flow is used for model training; after obtaining the intermediate result of the training model (completing one or several epochs) in the process, if the interaction frequency requirement determined above is met, the execution server sends the current model parameter (which can be encrypted by symmetric, asymmetric, homomorphic encryption, etc.) to the core server; the core server integrates the received execution server model parameters to obtain overall model parameters, and then the core server sends new overall model parameters (which can be encrypted by symmetric, asymmetric, homomorphic encryption, etc.) to each execution server; each execution server takes the new overall model parameters as a baseline and restarts the above modeling process until the model meets the corresponding standard and the modeling process is stopped.

[0081] The model parameter synthesis method process is: determining the execution server weight, first obtaining the number of training samples X (vector) on each execution server, then calculating the proportion of the number of training samples (SUM) of each execution server to the total number of training samples (0-1, Y-vector), according to the interval position of the proportion of the number of training samples of each execution server to the total number of training samples (such as 0-0.2 is set to 0.1, 0.2-0.4 is set to 0.3, etc.), the weight proportion of each execution server is set to Z (vector); Calculate the parameters of the integrated model, multiply the corresponding parameters in the model of each execution server by their respective weights, and then average to obtain.

[0082] For the high-performance model described above, input the content on the execution server into the model (including previous data processing calculations), and obtain the model output result (including executing other calculations).

[0083] In one embodiment, the method further comprises: calculating a data quality control index for the processing result of each link in the data analysis execution flow; the data quality control index is used to evaluate the usability of the data; and generating a result evaluation parameter for the final processing result.

[0084] Data quality control mainly evaluates the entire data analysis process (conducted along with the data analysis process), thereby providing process basis for reasonably viewing the result. In addition, the method can also include a model evaluation part, which mainly objectively evaluates the result from the data analysis execution itself.

[0085] That is, this step mainly includes data quality control, model evaluation and other parts. Among them, the model evaluation mainly objectively evaluates the result from the data analysis execution itself; and the data quality control mainly evaluates the entire data analysis process (conducted along with the data analysis process), thereby providing process quality basis for reasonably viewing the result.

[0086] The model evaluation here mainly provides different evaluation parameters to objectively understand the result itself. The evaluation parameters include significance level in statistical analysis, correlation coefficient size, regression coefficient size, accuracy, recall rate, F1, ROC, AUC, etc. of machine learning model.

[0087] The result of data quality control is a dimensionless number (for each data object, etc.), called data quality control index (0-100), which is divided into several ranges according to the size of the index: less than 60, the data is unusable (the data in this state will not participate in the data analysis process), 60-80, the data is usable, and greater than 80, the data is better.

[0088] The calculation process of the data quality control index, such as for structural images, first extracts feature data of the image data, including parameter differences (such as repetition time), signal-to-noise ratio, differences between structural images and templates, brain tissue symmetry, gray matter outside the template, and differences between different tool calculation results. After standardization and normalization of these data (standardization and normalization according to the pre-determined gold standard structural image and its numerical value), the data is weighted (the weight is determined in advance by the Delphi method) and normalized to 0-100.

[0089] The calculation results of data analysis, result evaluation, data quality control index, etc. are saved to the location specified by the user in the format (such as bids) mode determined in advance.

[0090] The above method provided by the embodiment combines medical image and secure computing features, and realizes a medical image data processing process ensuring data security through knowledge base construction, information configuration, data checking, data analysis flow determination, data analysis, result evaluation, and result acquisition, etc. Under the premise of data security, the efficiency of data processing and the effect of system modeling are improved.

[0091] After the data analysis, modeling evaluation, and data quality control data analysis calculation process is completed, the modeling model and results can be viewed, the model and results can be downloaded (as needed), and the model can be further applied through the graphical interface provided by the method.

[0092] Compared with the prior art, the medical image data processing method for realizing data security provided by the embodiment can one-stop select and automatically complete the whole process from preprocessing, feature extraction, statistical analysis, machine learning modeling, and deep learning modeling by configuring the target of medical image data processing and data related conditions through the interface.

[0093] As a specific example, in combination with Figure 2 The medical image data processing method for realizing data security provided by the embodiment specifically includes the following steps:

[0094] S210 knowledge base construction; to realize the neural image data processing calculation ensuring data security, a neural image data processing knowledge base is first constructed, and the framework thereof is constructed by using a multi-level multi-link method. The top-level categories of the knowledge base framework include: processing target, object data, calculation process, file, task, etc.

[0095] S220 information configuration; to maximize the convenience of users in neural image related data processing, the method provides a graphical interface for users to express neural image data processing related requirements. The requirement content for users to express data processing related in the graphical interface can be divided into four categories: data processing purpose, data itself (i.e. original data), data processing expectation, storage and server participating in calculation, etc.

[0096] S230 Data checking; in order to effectively implement the neural image data processing for data security, considering the possible differences in data of each data owner, it is necessary to first check and process the data. Mainly including data format conversion, data selection, data conversion, data anonymization, etc.

[0097] S240 Data analysis execution flow determination; according to the relevant requirements of users on processing data in the information configuration, including disease, focus and data situation, etc., based on the established data analysis decision model, the specific execution procedure of data analysis flow is obtained, including preprocessing, feature extraction, statistical analysis, machine learning modeling, deep learning modeling, result annotation, processing strategy, etc., and the interaction frequency parameters between the execution server and the core server based on the automatic recommendation of decision model generation.

[0098] S250 Data analysis; after the user customizes and confirms the data analysis execution flow, the data analysis execution flow (not including machine learning modeling and deep learning modeling) is executed on each execution server, and the data analysis process is automatically completed. Then each execution server starts to execute the system modeling process (machine learning modeling and deep learning modeling), and periodically interacts with the core server for model parameters until the model reaches the corresponding standard, and the modeling process is stopped.

[0099] S260 Modeling evaluation; this step mainly includes data quality control and model evaluation. Among them, model evaluation is mainly to objectively evaluate the results from data analysis execution itself; and data quality control is mainly to evaluate the whole process of data analysis (along with the data analysis process), so as to provide process basis for reasonably viewing the results.

[0100] S270 Model application; after the calculation and analysis process of data analysis, model evaluation and data quality control, the model and result can be viewed, downloaded (according to the need) and further applied by using the model through the graphical interface provided by the method.

[0101] The method can support automatic recommendation of a medical image (for example, a neural image) data analysis process and automatic recommendation of parameter setting of the analysis process based on an evidence-based medical knowledge base and a standardized literal sequence expression method, and can realize efficient calculation of the medical image (for example, the neural image) data on the basis of safe calculation, guarantee data security (the data is available but invisible to a non-data owner) while completing system modeling, and improve the effect of medical image data modeling and the generalization ability of a model by comprehensively using data characteristics distributed in different regions based on a safe calculation idea.

[0102] In addition, the embodiment of the present application further provides a medical image data processing system for realizing data security, which is applied to a server, and the server comprises an execution server and a core server. Figure 3 As shown in the figure, the system comprises a data generation module 310 configured to perform data checking and anonymization processing on original data obtained in advance to generate to-be-processed data.

[0103] A model prediction module 320 is configured to perform prediction by using a pre-generated decision model according to a user data processing requirement to generate a data analysis execution flow and an interaction frequency parameter respectively; the decision model comprises a data analysis decision model and an interaction frequency parameter automatic recommendation decision model; the interaction frequency parameter is used to represent an interaction frequency between the execution server and the core server; the data analysis execution flow comprises a plurality of medical image data analysis links and a corresponding order; the medical image data analysis link comprises a data processing link and a modeling link.

[0104] A data processing module 330 is configured to process the to-be-processed data by the execution server based on the data analysis execution flow to complete data analysis of the data processing link.

[0105] A result generation module 340 is configured to construct a corresponding model based on the data analysis execution flow to generate a server comprehensive model to complete model construction of the modeling link; the server comprehensive model is used to perform prediction on input data and generate a final processing result in combination with the data processing link to realize automatic and safe analysis and processing of the medical image data.

[0106] The medical image data processing system for realizing data security provided in the embodiment of the present application can be specific hardware on a device or software or firmware installed on the device, and the device provided in the embodiment of the present application has the same implementation principle and technical effects as the foregoing method embodiment.

[0107] The device embodiment part is not mentioned, please refer to the corresponding content in the foregoing method embodiment. Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the foregoing described system, device and unit can refer to the corresponding process in the above method embodiment, which will not be repeated here. The medical image data processing system for realizing data security provided by the embodiments of the application has the same technical features as the medical image data processing method for realizing data security provided by the above embodiments, so it can also solve the same technical problems and achieve the same technical effects.

[0108] 0The embodiments of the present application also provide an electronic device, specifically, the electronic device includes a processor and a storage device; the storage device stores a computer program, and the computer program executes the method according to any one of the above embodiments when being run by the processor.

[0109] Figure 4 The structure schematic diagram of an electronic device provided by the embodiments of the present application is shown in the figure, and the electronic device 400 includes a processor 40, a memory 41, a bus 42 and a communication interface 43, the processor 40, the communication interface 43 and the memory 41 are connected through the bus 42; the processor 40 is used to execute the executable module stored in the memory 41, such as a computer program.

[0110]

[0111] Wherein, the memory 41 can contain a high-speed random access memory (RAM, Random Access Memory), and can also include a non-volatile memory, such as a magnetic disk memory.

[0112] The system network element and at least one other network element are connected through at least one communication interface 43 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used.

[0113]

[0114] The bus 42 can be an ISA bus, a PCI bus or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one bidirectional arrow is used, but it does not mean that there is only one bus or one type of bus. Figure 4

[0115] Wherein, the memory 41 is used to store programs, and the processor 40 executes the programs after receiving execution instructions. The method executed by the device defined by the flow process disclosed in any one of the above embodiments can be applied to the processor 40 or realized by the processor 40. ​​​

[0116] The processor 40 can be an integrated circuit chip with signal processing capability. In implementation, each step of the above method can be completed by integrated logic circuit of hardware in the processor 40 or by instructions in the form of software. The processor 40 described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; or can be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. Each method, step and logic block diagram disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as a hardware coding processor for execution, or a combination of hardware and software modules in the coding processor for execution. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium in the memory 41 is read by the processor 40, and the hardware thereof is combined to complete the steps of the above method.

[0117] Corresponding to the above method, the embodiments of the present application also provide a computer readable storage medium, the computer readable storage medium stores machine executable instructions, when the processor calls and runs the computer executable instructions, the computer executable instructions cause the processor to run the steps of the above method.

[0118] In the embodiments provided by the present application, it should be understood that the disclosed device and method can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection can be indirect coupling or communication connection through some communication interface, device or unit, and can be electrical, mechanical or other forms.

[0119] The unit described as a separate component may or may not be physically separate; the component shown as a unit may or may not be a physical unit, that is, it may be located in one place.

[0120] Alternatively, the units can be distributed across multiple network units. Some or all of these units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0121] Furthermore, the functional units in the embodiments provided in this application can be integrated into a single processing unit.

[0122] In a unit, each unit can exist physically independently, or two or more units can be integrated into a single unit.

[0123] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a portion of the technical solution, is essentially...

[0124] The components can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] It should be noted that similar labels and letters in the accompanying drawings indicate similar items. Therefore, once an item is defined in one accompanying drawing, it does not need to be further defined and explained in subsequent accompanying drawings. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that...

[0127] Those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for processing medical image data to achieve data security, characterized in that, The method is applied to a server, which includes an execution server and a core server; the method includes: The pre-acquired raw data is inspected and anonymized to generate data to be processed; The system acquires predefined rules between interaction frequency parameters, user data processing requirements, the number of servers participating in the calculation, and the data analysis execution flow; these predefined rules are used as the basis for relevant decisions to recommend interaction frequency parameters. Based on relevant data during server operation, a decision recommendation model is constructed using machine learning methods; the inputs to the decision recommendation model are the encoded user data processing requirements, the number of servers, and the data analysis execution flow; the output of the decision recommendation model is the interaction frequency parameter. Based on the predefined rules and the decision recommendation model, an automatic recommendation decision model is determined by determining the interaction frequency parameter; Based on user data processing requirements, predictions are made using a pre-generated decision model, generating a data analysis execution flow and interaction frequency parameters. The decision model includes a data analysis decision model and an automatic recommendation decision model for the interaction frequency parameters. The interaction frequency parameters represent the interaction frequency between the execution server and the core server. The data analysis execution flow includes several medical image data analysis steps and their corresponding order. Each medical image data analysis step includes a data processing step and a modeling step. Based on the data analysis execution flow, the execution server processes the data to be processed to complete the data analysis in the data processing stage; Based on the data analysis execution flow, a corresponding model is constructed to generate a server comprehensive model, thereby completing the model construction in the modeling stage. The server comprehensive model is used to predict the input data and, in conjunction with the data processing stage, generate the final processing result to achieve automatic and secure analysis and processing of medical image data.

2. The medical image data processing method for achieving data security according to claim 1, characterized in that, The steps of performing data inspection and anonymization on the pre-acquired raw data to generate the data to be processed include: The pre-acquired raw data is checked according to a predefined data format; if the raw data does not conform to the predefined data format, the raw data is converted according to the predefined data format. The data attributes of the raw data are checked to determine whether the raw data meets the analysis conditions; The raw data under different acquisition parameters are processed to convert the raw data into standard data; The header information and facial feature information of the standard data are de-privatized to generate data to be processed.

3. The medical image data processing method for achieving data security according to claim 1, characterized in that, The user data processing requirements include: data processing purpose, raw data, data processing expectations, and storage and servers involved in the computation; Before the steps of generating data analysis execution flow and interaction frequency parameters based on user data processing requirements and using a pre-generated decision model for prediction, the method further includes: A training database is generated based on a pre-built medical image data processing knowledge base. The framework of the medical image data processing knowledge base includes multiple primary categories. The primary categories include: processing target, object data, analysis process, file, and task. The data analysis decision model is trained using the training database to generate a trained data analysis decision model.

4. The medical image data processing method for achieving data security according to claim 1, characterized in that, Based on user data processing requirements, the steps of generating data analysis execution flow and interaction frequency parameters by using a pre-generated decision model for prediction include: Based on user data processing requirements, a pre-trained data analysis decision model is used to make predictions and generate a data analysis execution flow. The interaction frequency parameters are used to automatically recommend a decision-making model for prediction, thereby generating the interaction frequency parameters.

5. The medical image data processing method for achieving data security according to claim 1, characterized in that, The steps for building a comprehensive server model based on the data analysis execution flow include: Each execution server performs corresponding model training for the modeling stage. If the training process meets the interaction frequency parameter requirements, it sends the current model parameters to the core server. The core server generates comprehensive model parameters based on the received current model parameters, and sends the comprehensive model parameters to each of the execution servers; Each execution server trains based on the parameters of the comprehensive model, repeats the parameter interaction process until the current model reaches the corresponding standard, and determines the current model as the server comprehensive model.

6. The medical image data processing method for achieving data security according to any one of claims 1 to 5, characterized in that, The method further includes: For the processing results of each stage in the data analysis execution flow, a data quality control index is calculated respectively; the data quality control index is used to evaluate the usability of the data. Result evaluation parameters are generated based on the final processing result.

7. A medical image data processing system for achieving data security, characterized in that, The system is applied to a server, which includes an execution server and a core server; the system includes: The data generation module is used to perform data inspection and anonymization on the pre-acquired raw data to generate data to be processed; A data acquisition module is used to acquire interaction frequency parameters, user data processing requirements, the number of servers participating in the calculation, and predefined rules between the data analysis execution flow. These predefined rules serve as the basis for recommending interaction frequency parameters for relevant decisions. Based on relevant data during server operation, a decision recommendation model is constructed using machine learning methods. The input to the decision recommendation model is the encoded user data processing requirements, the number of servers, and the data analysis execution flow. The output of the decision recommendation model is the interaction frequency parameters. Based on the predefined rules and the decision recommendation model, the interaction frequency parameters are determined to automatically recommend a decision model. The model prediction module is used to make predictions based on user data processing requirements using a pre-generated decision model, and to generate data analysis execution flow and interaction frequency parameters respectively. The decision model includes a data analysis decision model and an automatic recommendation decision model based on interaction frequency parameters. The interaction frequency parameters represent the interaction frequency between the execution server and the core server. The data analysis execution flow includes several medical image data analysis steps and their corresponding order. The medical image data analysis steps include a data processing step and a modeling step. The data processing module is used to process the data to be processed by the execution server based on the data analysis execution flow, so as to complete the data analysis of the data processing stage; The result generation module is used to build a corresponding model based on the data analysis execution flow and generate a server comprehensive model to complete the model building in the modeling stage. The server comprehensive model is used to predict the input data and generate the final processing result in combination with the data processing stage to realize the automatic and secure analysis and processing of medical image data.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores machine-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method according to any one of claims 1 to 6.

Images