Search

CN-122024980-A - Medical intelligent calculation model service platform and implementation method thereof

CN122024980ACN 122024980 ACN122024980 ACN 122024980ACN-122024980-A

Abstract

The invention relates to the technical field of data management, in particular to a medical intelligent computing model service platform and an implementation method thereof. The method comprises the steps of collecting a multi-source historical medical data set, executing structure consistency processing to obtain a basic data set, calculating feature correlation and distribution difference between modes based on the basic data set, establishing feature mapping relations corresponding to the modes, determining node safety boundary ranges of data sources by means of the feature mapping relations, inputting the feature mapping relations and the basic data set into a medical intelligent computing model frame to train a medical intelligent computing model, dividing the medical intelligent computing model into a plurality of functional containers, setting encryption cooperation mechanisms between the functional containers and data source links according to the node safety boundary ranges, and uploading the functional containers with the encryption cooperation mechanisms to a cloud platform to execute containerized deployment. The invention enhances the reliability and overall security of medical data processing.

Inventors

  • Kou Liyan

Assignees

  • 和光数字(深圳)有限公司

Dates

Publication Date
20260512
Application Date
20260113

Claims (10)

  1. 1. The implementation method of the medical intelligent computing model service platform is characterized by comprising the following steps of: Step S1, collecting a multi-source historical medical data set, and executing structure consistency processing on the multi-source historical medical data to obtain a basic data set; step S2, calculating the feature correlation degree and the distribution difference degree among the modes based on the feature vector of each mode in the basic data set, and establishing a feature mapping relation corresponding to each mode; Step S3, performing feature risk classification on each data source in the basic data set by utilizing the feature mapping relation, and determining the node safety boundary range of each data source according to the feature risk classification result; S4, inputting the feature mapping relation and the basic data set into a medical intelligent computing model frame to train the medical intelligent computing model; And S5, dividing the medical intelligent computation model into a plurality of functional containers, setting encryption cooperative mechanisms between each functional container and the data source link according to the security boundary range of the node, and uploading the functional containers with the encryption cooperative mechanisms to the cloud platform to execute containerized deployment.
  2. 2. The method for implementing a medical intelligent computing model service platform according to claim 1, wherein after constructing the medical intelligent computing model service platform, further comprises: receiving a data access request of a data source, and determining a functional container matched with the data access request through a medical intelligent computing model service platform; Determining the credibility of the data access request according to an encryption cooperative mechanism between the matched functional container and the data source link; And linking a data source corresponding to the data access request with the reliability higher than a preset reliability threshold with the matched functional container, and dynamically scheduling corresponding computing resources according to the current resource allocation condition of the medical intelligent computing model service platform so as to execute data access and model reasoning tasks.
  3. 3. The method for implementing a medical intelligent computing model service platform according to claim 2, wherein after performing the data access and model reasoning tasks, further comprises: acquiring user feedback data returned by a data source, and comparing model reasoning path deviation and reasoning delay deviation between the user feedback data and a model reasoning result; if the model reasoning path deviation exceeds a preset path deviation threshold, performing iterative training and parameter updating on the medical intelligent calculation model until the model reasoning path deviation is in the path deviation threshold; If the reasoning delay deviation exceeds a preset delay deviation threshold, according to the current resource allocation situation of the medical intelligent computing model service platform, adjusting the computing resources corresponding to the matched functional containers so as to re-execute the data access and model reasoning tasks.
  4. 4. The implementation method of the medical intelligent computing model service platform according to claim 1, wherein the structure unification processing in step S1 includes: Performing format recognition on the multi-source historical medical data set, and dividing the multi-source historical medical data set into text medical record data, image frame data and physiological monitoring data; performing pixel blocking on the image frame data, and extracting edge features according to pixel blocking results; And (3) hierarchically and structurally extracting text semantic descriptions in the text medical record data as a main mode semantic tag, and carrying out semantic association matching on edge features and the noise-removed physiological monitoring data according to the main mode semantic tag to obtain a basic data set.
  5. 5. The implementation method of the medical intelligent computing model service platform according to claim 1, wherein the calculating of the feature correlation and the distribution difference between the modalities in step S2 includes: According to the feature vector of each mode in the basic data set, calculating the relative offset of the feature distribution center between any two modes, and taking the relative offset as the distribution difference degree; Matching sample characteristics of each mode in the basic data set under the same time window, and calculating a joint variance and a covariance value between the matched sample characteristic pairs so as to construct an inter-mode covariance matrix; And calculating the feature correlation degree among the modes based on the inter-mode covariance matrix, and if the feature correlation degree among the modes is lower than a preset correlation degree threshold value, re-selecting a sample feature pair consistent with the semantic label of the main mode from the corresponding modes with the feature correlation degree lower than the correlation degree threshold value, and re-calculating the feature correlation degree until the feature correlation degree among the modes is larger than the correlation degree threshold value.
  6. 6. The implementation method of the medical intelligent computing model service platform according to claim 1, wherein the establishing of the feature mapping relation corresponding to each mode in step S2 includes: Taking a modal pair with the feature correlation higher than a correlation threshold as a candidate mapping pair, and executing linear translation correction based on the distribution difference of the candidate mapping pair until the distribution difference of the candidate mapping pair is smaller than a preset difference threshold to obtain a distribution alignment modal pair; Extracting feature correlation of distribution alignment mode pairs based on the inter-mode covariance matrix, and establishing a mapping weight function between modes according to the feature correlation; Projecting the edge features and the physiological monitoring data in the basic data set to the corresponding semantic tag ranges in the semantic tags of the main modes on a time axis, and adjusting the projection positions based on the mapping weight function so as to obtain the feature mapping relation corresponding to each mode.
  7. 7. The method for implementing a medical intelligent computing model service platform according to claim 1, wherein the performing feature risk classification in step S3 includes: detecting the value fluctuation range of each feature based on the feature mapping relation, and marking the feature with the value exceeding the historical value range of the main mode semantic tag as an abnormal candidate feature; according to the consistency of the value exceeding amplitude, the occurrence frequency and the value of the related mode pair of the abnormal candidate feature on the time sequence, and according to a preset weighting proportion, weighting calculation is carried out to obtain a feature risk index; And executing interval layering according to the characteristic risk index distribution condition, and taking an interval layering result as a characteristic risk grading result.
  8. 8. The implementation method of the medical intelligent computing model service platform according to claim 1, wherein determining the node security boundary range of each data source in step S3 includes: Detecting data interaction nodes of all data sources in a cloud platform, and extracting node behavior characteristics of the data interaction nodes; Comparing the node behavior characteristics with characteristic risk classification results in corresponding time windows, and calculating node risk response rate; comparing the risk response rate of the nodes with a preset response rate threshold value to divide the risk level of the nodes; And determining the risk propagation path intensity among the nodes by using the node risk response rate, and dividing the node safety boundary range of each data source according to the risk propagation path intensity and the risk grade of each node.
  9. 9. The implementation method of the medical intelligent computing model service platform according to claim 1, wherein the setting of the encryption collaboration mechanism between each functional container and the data source link in step S4 includes: Determining a link relation between the functional container and the data source, and selecting an encryption strategy corresponding to the link relation according to the node safety boundary range; And selecting a cooperative strategy corresponding to the link relation according to the sensitivity of the data source corresponding to the acquired data and the sensitivity of the reasoning result of the functional container, and setting the communication channels of the functional containers according to the encryption strategy and the cooperative strategy.
  10. 10. A medical intelligent computing model service platform, for executing the medical intelligent computing model service platform implementation method of claim 1, the medical intelligent computing model service platform comprising: the data acquisition module is used for acquiring a multi-source historical medical data set and executing structure consistency processing on the multi-source historical medical data to obtain a basic data set; the feature analysis module is used for calculating the feature correlation degree and the distribution difference degree among the modes based on the feature vectors of the modes in the basic data set and establishing a feature mapping relation corresponding to the modes; The boundary determining module is used for executing feature risk classification on each data source in the basic data set by utilizing the feature mapping relation and determining the node safety boundary range of each data source according to the feature risk classification result; the model training module is used for inputting the characteristic mapping relation and the basic data set into the medical intelligent computing model frame so as to train the medical intelligent computing model; The function container deployment module is used for dividing the medical intelligent computation model into a plurality of function containers, setting encryption cooperation mechanisms between each function container and the data source link according to the node safety boundary, and uploading the function containers with the encryption cooperation mechanisms to the cloud platform to execute containerized deployment.

Description

Medical intelligent calculation model service platform and implementation method thereof Technical Field The invention relates to the technical field of data management, in particular to a medical intelligent computing model service platform and an implementation method thereof. Background The medical data comprise multi-source heterogeneous information such as electronic medical records, image data, genome data, wearable equipment data and various inspection indexes. The data has the characteristics of high dimensionality, strong time sequence, complex structure, privacy sensitivity and the like. The existing medical data processing and analyzing modes mainly depend on a single algorithm or a traditional statistical method, and the isomerism, the mass property and the real-time property of the data are difficult to effectively support. The existing intelligent medical service platform has the defects that on one hand, the platform usually depends on a discrete model or an isolated algorithm, a unified computing frame crossing data sources and tasks is lacked, so that the model training, deployment and updating efficiency is low, and the model cannot adapt to the dynamic change of a clinical simulation scene, on the other hand, the problems of data safety, privacy protection and compliance restrict the data sharing and collaborative computing of multiple mechanisms and multiple scenes, and the model performance is prevented from being improved. Disclosure of Invention Accordingly, the present invention is directed to a medical intelligent computing model service platform and a method for implementing the same, which solve at least one of the above-mentioned problems. In order to achieve the above purpose, a method for implementing a medical intelligent computation model service platform comprises the following steps: Step S1, collecting a multi-source historical medical data set, and executing structure consistency processing on the multi-source historical medical data to obtain a basic data set; step S2, calculating the feature correlation degree and the distribution difference degree among the modes based on the feature vector of each mode in the basic data set, and establishing a feature mapping relation corresponding to each mode; Step S3, performing feature risk classification on each data source in the basic data set by utilizing the feature mapping relation, and determining the node safety boundary range of each data source according to the feature risk classification result; S4, inputting the feature mapping relation and the basic data set into a medical intelligent computing model frame to train the medical intelligent computing model; And S5, dividing the medical intelligent computation model into a plurality of functional containers, setting encryption cooperative mechanisms between each functional container and the data source link according to the security boundary range of the node, and uploading the functional containers with the encryption cooperative mechanisms to the cloud platform to execute containerized deployment. The multi-mode feature mapping and risk classification method has the advantages that multi-mode fusion and intelligent analysis of text medical records, image frames and physiological monitoring data are realized through structural processing, feature mapping and risk classification of multi-source heterogeneous medical data, data with different sources and different formats can be effectively integrated, model training and reasoning efficiency is remarkably improved, after the model is divided into functional containers and combined with node safety boundaries and encryption cooperative mechanisms, a platform can realize multi-mechanism and multi-scene data access and cooperative calculation on the premise of guaranteeing data safety and privacy, data leakage risk and compliance cost are reduced, the model performance can be continuously optimized through dynamic scheduling of calculation resources and combined with a user feedback reasoning path deviation and delay deviation optimization mechanism, diagnosis and prediction accuracy and response speed are improved, in addition, the multi-mode feature mapping and risk control mechanism can identify and isolate abnormal features and high-risk nodes in real time, reliability of medical data processing and overall safety of a system are improved, and intelligent computing capability of medical intelligence, safety and sustainable optimization is realized as a whole. Optionally, the present application provides a medical intelligent computing model service platform for executing the medical intelligent computing model service platform implementation method as described above, where the medical intelligent computing model service platform includes: the data acquisition module is used for acquiring a multi-source historical medical data set and executing structure consistency processing on the multi-source historical medical data