Cross-modal data generation-driven chronic disease management method and platform

Abstract

The present invention discloses a chronic disease management method and platform driven by cross-modal data generation, which relates to the technical field of medical data management. The method includes: accessing a cross-modal database; obtaining multiple information receiving ends, and extracting feature vectors representing data granularity requirements; outputting clustering results, configuring the number of multiple transmission channels, and establishing connections between the multiple transmission channels and the multiple information receiving ends; setting multiple data granularities of the multiple transmission channels, performing template processing, and performing templatization processing on the chronic disease resource data transmitted by the multiple transmission channels according to multiple granularity templates, and correspondingly transmitting them to the multiple information receiving ends. It solves the technical problems in the prior art that the cross-modal data granularity is inconsistent with the modal semantics, and it is difficult to flexibly generate appropriate granularity templates according to requirements, resulting in low information transmission efficiency and poor accuracy and timeliness of chronic disease management, and achieves the technical effect of improving the data transmission efficiency and the accuracy and timeliness of chronic disease management.

Description

Technical Field

[0001] This application relates to the technical field of medical data management, and particularly to a chronic disease management method and platform driven by cross-modal data generation. Background Art

[0002] In the management of chronic diseases, since patients with chronic diseases require long-term and multi-dimensional health data tracking and monitoring, traditional medical health management faces problems such as heterogeneous sources of multi-modal data, low data processing efficiency, and inaccurate information transmission. The management of chronic diseases not only needs to process different types of data, such as medical images, laboratory test data, vital sign monitoring data, data self-reported by patients, etc., but also the data distribution, format, dimension, etc. of each modality vary greatly, resulting in difficulties in information fusion and cross-modal analysis; it also needs to generate data templates suitable for different granularities according to different information needs to better support decision-making and early warning. The existing single data processing mode cannot meet these diverse needs, and the information granularity requirements in different chronic disease management scenarios are different, making it difficult to flexibly generate appropriate granularity templates according to requirements, resulting in low data transmission and display efficiency, and even affecting the accuracy of decision-making and the accuracy of chronic disease management.

[0003] Therefore, in the related technologies at the present stage, there are technical problems such as inconsistent cross-modal data granularity and modal semantics, difficulty in flexibly generating appropriate granularity templates according to requirements, resulting in low information transmission efficiency, and poor accuracy and real-time performance of chronic disease management. Summary of the Invention

[0004] By providing a chronic disease management method and platform driven by cross-modal data generation, this application solves the technical problems in the prior art, such as inconsistent cross-modal data granularity and modal semantics, difficulty in flexibly generating appropriate granularity templates according to requirements, resulting in low information transmission efficiency, and poor accuracy and real-time performance of chronic disease management, and achieves the technical effect of improving data transmission efficiency, and the accuracy and real-time performance of chronic disease management.

[0005] The present application provides a chronic disease management method driven by cross-modal data generation, and the method includes: accessing a cross-modal database, where the cross-modal database is stored in a cloud processor, and the cloud processor includes multiple transmission channels; obtaining multiple information receivers, and extracting a feature vector representing the data granularity requirement in the multiple information receivers; clustering the multiple information receivers according to the feature vector, outputting a clustering result, configuring the number of the multiple transmission channels according to the number of classes of the clustering result, and establishing connections between the multiple transmission channels and the multiple information receivers; setting multiple data granularities of the multiple transmission channels, respectively performing template processing according to the multiple data granularities, outputting multiple granularity templates, downloading the multiple granularity templates into the multiple transmission channels, and performing templating processing on the chronic disease resource data transmitted by the multiple transmission channels, and correspondingly transmitting it to the multiple information receivers.

[0006] In a possible implementation manner, the chronic disease management method driven by cross-modal data generation further performs the following processing: performing high-dimensional feature space mapping on the data of each modality in the cross-modal database, and outputting a multi-modal high-dimensional feature data set corresponding to multiple modalities; using a functional adversarial network to perform generative alignment on the multi-modal high-dimensional feature data set, and outputting an aligned multi-modal feature data set; using the aligned multi-modal feature data set to perform modal data feature fusion, and updating the cross-modal database.

[0007] In a possible implementation manner, the chronic disease management method driven by cross-modal data generation further performs the following processing: where the functional adversarial network includes a generator and a discriminator, the output end of the generator is connected to the input end of the discriminator, and the output end of the discriminator is connected to the input end of the generator; performing adversarial training on the generator and the discriminator, obtaining the modal data generated by the generator, inputting the generated modal data into the discriminator, calculating the difference distribution from the sample modal data, and performing parameter learning with the goal of being less than a preset difference distribution to obtain a trained functional adversarial network.

[0008] In a possible implementation manner, the chronic disease management method driven by cross-modal data generation further performs the following processing: calculating the semantic difference degree between the features in each modality of the aligned multi-modal feature data set; storing the features with a semantic difference degree greater than or equal to the preset semantic difference degree in parallel, and storing the features with a semantic difference degree less than the preset semantic difference degree in a covering manner.

[0009] In a possible implementation, the cross-modal data generative-driven chronic disease management method further performs the following processing: The cloud processor performs templatization processing on the cross-modal database to generate an initialization template, where the initialization template includes the number of initialization template partitions, the initialization information hierarchy structure, the initialization data update frequency, and the initialization data display accuracy; adjust the template parameters of the initialization template according to the multiple data granularities, and output multiple granularity templates, where the template parameters include the number of template partitions, the information hierarchy structure data, the update frequency, and the data display accuracy.

[0010] In a possible implementation, the cross-modal data generative-driven chronic disease management method further performs the following processing: The multi-party information receiver performs quality assessment on the received templatized chronic disease resource data to obtain multiple quality assessment indicators corresponding to the multi-party information receiver; feedback the multiple quality assessment indicators to the cloud processor, and update the template parameters of the multiple granularity templates according to the cloud processor until the preset quality assessment indicators are met, and output the updated multiple granularity templates.

[0011] In a possible implementation, the cross-modal data generative-driven chronic disease management method further performs the following processing: Use the multi-party information receiver as an edge node and the cloud processor as a central node to construct a resource topology management platform; perform update frequency synchronization processing on the granularity templates of the multiple transmission channels based on the resource topology management platform.

[0012] In a possible implementation, the cross-modal data generative-driven chronic disease management method further performs the following processing: where the feature vector of each information receiver includes data fineness requirements, data interaction complexity, data modality preference, and granularity level requirements; calculate the similarity between the feature vectors of every two information receivers among the multi-party information receivers to obtain a similarity set; use the similarity set to cluster the multi-party information receivers, and output a clustering result.

[0013] The present application also provides a chronic disease management platform driven by cross-modal data generation, including: a cross-modal database access unit for accessing a cross-modal database stored in a cloud processor, where the cloud processor includes multiple transmission channels; a feature vector extraction unit for obtaining multiple information receiving ends and extracting feature vectors representing data granularity requirements in the multiple information receiving ends; a transmission channel configuration unit for clustering the multiple information receiving ends according to the feature vectors, outputting a clustering result, configuring the number of the multiple transmission channels according to the number of classes in the clustering result, and establishing connections between the multiple transmission channels and the multiple information receiving ends; a granularity template output unit for setting multiple data granularities of the multiple transmission channels, performing template processing according to the multiple data granularities respectively, outputting multiple granularity templates, downloading the multiple granularity templates into the multiple transmission channels, and performing templatization processing on the chronic disease resource data transmitted by the multiple transmission channels according to the multiple granularity templates and transmitting them to the multiple information receiving ends correspondingly.

[0014] It is intended to access a cross-modal database through the chronic disease management method and platform driven by cross-modal data generation proposed in the present application; obtain multiple information receiving ends, extract feature vectors representing data granularity requirements; cluster the multiple information receiving ends, output a clustering result, configure the number of multiple transmission channels, and establish connections between the multiple transmission channels and the multiple information receiving ends; set multiple data granularities of the multiple transmission channels, perform template processing, perform templatization processing on the chronic disease resource data transmitted by the multiple transmission channels according to multiple granularity templates, and transmit them to the multiple information receiving ends correspondingly. The technical problems existing in the prior art, such as inconsistent cross-modal data granularity and modal semantics, and difficulty in flexibly generating appropriate granularity templates according to requirements, resulting in low information transmission efficiency and poor accuracy and real-time performance of chronic disease management, are solved, and the technical effects of improving data transmission efficiency and the accuracy and real-time performance of chronic disease management are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in the present application to illustrate the operations performed by the platform according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be executed precisely in sequence. On the contrary, according to needs, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0016] Figure 1 It is a schematic flowchart of a chronic disease management method driven by cross-modal data generation provided by an embodiment of the present application.

[0017] Figure 2Schematic diagram of the chronic disease management platform driven by cross-modal data generation provided by the embodiments of the present application.

[0018] Explanation of reference numerals: Cross-modal database access unit 10, feature vector extraction unit 20, transmission channel configuration unit 30, granularity template output unit 40. Detailed implementation manners

[0019] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specifically gives the detailed implementation manners of the present application.

[0020] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0021] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first / second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, platform, product or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0022] The embodiments of the present application provide a chronic disease management method driven by cross-modal data generation, as Figure 1 shown, the method includes:

[0023] Step S100, accessing a cross-modal database, where the cross-modal database is stored in a cloud processor, and the cloud processor includes a plurality of transmission channels.

[0024] Preferably, a cross-modal database is used to store different types of modal data. Here, modal data refers to data from different sources and forms, including but not limited to medical imaging data, electronic health records (EHRs), laboratory test results, sensor data, patients' self-reported data, etc. The cross-modal database can uniformly manage data from different modalities (data with different formats and characteristics) and support their integration and analysis. The cross-modal database is stored in a cloud processor, which means that all modal data is centrally stored on a cloud server, providing efficient computing resources and storage space for data access, processing, and analysis. The cloud processor has strong computing capabilities, can support the processing of large-scale data, and can achieve real-time analysis and response of cross-modal data. Moreover, the cloud processor includes multiple different paths or channels for data transmission. Each channel is used for the transmission of different granularity or different types of data. Different transmission channels can be optimized for different data requirements (such as different client ends or different information display levels). Each transmission channel can process different data granularities, or adjust the format, type, or processing ability of data transmission according to user needs, which helps to flexibly transmit and process different data streams according to specific situations, thereby improving the platform response ability and information transmission efficiency, and achieving more refined and efficient cross-modal data management.

[0025] Further, step S100 further includes step S110, performing high-dimensional feature space mapping on the data of each modality in the cross-modal database, and outputting a multi-modal high-dimensional feature data set corresponding to multiple modalities; step S120, using a functional-adversarial network to perform generative alignment on the multi-modal high-dimensional feature data set, and outputting an aligned multi-modal feature data set; step S130, using the aligned multi-modal feature data set to perform modal data feature fusion and updating the cross-modal database.

[0026] Preferably, perform high-dimensional feature space mapping on various different modalities of medical data stored in the database (such as medical imaging data, clinical examination data, symptoms self-reported by patients, sensor monitoring data, etc.), that is, convert these different modalities of data into representations in a high-dimensional feature space. Data of each modality may have different structures and dimensions. For example, image data is a two-dimensional or three-dimensional matrix, text data may be a sequence, and sensor data is a time series, etc. By mapping to a high-dimensional feature space, data from different sources (such as images, time series, physiological signals, etc.) can share the same space, and then output the corresponding multi-modal high-dimensional feature data set.

[0027] Preferably, a functional adversarial network is used to perform generative alignment on the multi-modal high-dimensional feature dataset, that is, the high-dimensional feature data of multiple modalities are aligned so that the data of these modalities not only match in structure but also remain consistent in semantic and statistical features. Through the functional adversarial network, virtual data with the same semantics as the original modal data can be generated to ensure alignment between different modalities. Among them, the functional adversarial network is a method based on the generative adversarial network. By training two neural networks (a generator and a discriminator), data is generated in the generator, and the discriminator evaluates the difference between the generated data and the real data. The functional adversarial network focuses on data alignment at the functional level and can solve the problem of differences between modalities by generating data with the same distribution as the original data; then, the features in the aligned multi-modal data are fused, that is, the data of different modalities are weighted averaged, concatenated, etc. in the same space so that the fused data can retain the unique information of different modal data at the same time, reduce redundant information, and find the optimal combination method between multiple modalities; finally, the high-dimensional feature dataset obtained after generative alignment and feature fusion is inserted into the original cross-modal database as a new data record, thereby updating the content in the modal database and improving the accuracy and consistency of cross-modal data analysis.

[0028] Further, step S120 further includes step S121, where the functional adversarial network includes a generator and a discriminator, the output end of the generator is connected to the input end of the discriminator, and the output end of the discriminator is connected to the input end of the generator; step S122, perform adversarial training on the generator and the discriminator, obtain the modal data generated by the generator, input the generated modal data into the discriminator, calculate the difference distribution from the sample modal data, and perform parameter learning with the goal of being less than the preset difference distribution to obtain the trained functional adversarial network.

[0029] Preferably, the functional adversarial network (F-GAN) is an extension of the traditional generative adversarial network (GAN), which generally includes two core components, a generator and a discriminator. The generator generates fake data, and the discriminator determines the authenticity of the data. The two components compete with each other, gradually optimizing the ability of the generator so that the generated data becomes increasingly close to the real data. In the functional adversarial network (F-GAN), the design objectives of the generator and the discriminator are to solve the problems of cross-modal data generation and alignment. The generator not only generates "fake" data but also needs to generate outputs that match the target-modal data in a specific functional space, while the discriminator needs to evaluate the differences between the generated data and the real data based on the objective of "functional alignment"; the output end of the generator is connected to the input end of the discriminator. The modal data generated by the generator (e.g., synthetic medical images or text data) is input into the discriminator, and the task of the discriminator is to determine whether these data match the real data; the output end of the discriminator (i.e., the score or difference measurement result of the discriminator) is returned to the generator to guide the optimization of the generator. The generator adjusts its parameters according to the output of the discriminator, making the generated data closer to the distribution of the target-modal data.

[0030] Preferably, the generator and the discriminator are adversarially trained, that is, through the game between the generator and the discriminator, the generator continuously generates data that better meets the expected goals, while the discriminator continuously improves its recognition ability. During the training process, the generator will gradually adjust its generation strategy according to the feedback of the discriminator to minimize the difference between the generated data and the real data. The discriminator learns how to make effective judgments in the functional space by identifying the differences between the generated data and the real data. Specifically, at the beginning of the training, the generator receives inputs (such as random noise, partial information of the target modality, etc.) and generates modal data (medical images or other types of health data). The discriminator receives the modal data output by the generator and the real modal data (from the original samples or cross-modal databases), calculates the difference distribution between the two, and conducts parameter learning with the goal of being less than the preset difference distribution. For example, a certain distance metric or discriminant criterion is used to measure the similarity or difference degree, and then the generator parameters are adjusted through an optimization algorithm (such as gradient descent) to make the generated data as close as possible to the real data in terms of the difference distribution. The goal of the discriminator is to more accurately evaluate the differences between the generated data and the real data. Throughout the process, the generator and the discriminator continuously play games and reach a balance through adversarial training. The generated data by the generator is highly consistent with the real data in the functional space, and it is ensured that the generated data can meet certain functional requirements, such as semantic consistency or the alignment goal of cross-modal data; finally, the generator and the discriminator together form a functional adversarial network. The generator can generate modal data that meets the functional goals according to the input data, and the discriminator can evaluate the differences between these generated data and the real data.

[0031] Furthermore, step S130 further includes step S131, which calculates the semantic difference degree between features in each modality of the aligned multi-modal feature dataset; step S132 stores the features with a semantic difference degree greater than or equal to a preset semantic difference degree in parallel, and stores the features with a semantic difference degree less than the preset semantic difference degree by overwriting.

[0032] Preferably, the semantic difference degree between features in each modality of the aligned multi-modal feature dataset is calculated. The semantic difference degree is a metric for measuring the similarity or difference degree of different modality features (or different features in the same modality) at the semantic level. Specifically, the semantic difference degree is measured by calculating a certain distance between features (such as Euclidean distance, cosine similarity, etc.), which reflects the similarity of features in the semantic space. For example, it is represented by a similarity metric (such as cosine similarity) or a distance metric (such as L2 norm). A smaller semantic difference degree means that two features are more similar semantically, while a larger semantic difference degree means that two features are more different semantically, indicating that they may belong to different categories or come from different modalities in terms of representation.

[0033] Preferably, the preset semantic difference degree refers to a preset semantic difference degree setting threshold, which determines which features are stored in parallel and which features are stored by overwriting. Specifically, two features with a semantic difference degree greater than or equal to the preset semantic difference degree are stored in parallel, that is, these two features will retain their respective storage positions to ensure that they can be used separately in subsequent processing. For example, medical images and gene data may have a large semantic difference in the feature space and should be stored in parallel; if the semantic difference degree between two features is less than the preset threshold, they are stored by overwriting, that is, these features will be stored in the same position, overwriting the original feature representation, indicating that they are highly consistent semantically. For example, the feature data generated by two sensors (such as a blood glucose meter and a blood pressure monitoring device) can be stored by overwriting if their difference in the semantic space is small; thus, the storage efficiency of cross-modal data is improved, and the data processing flow of the entire chronic disease management is optimized.

[0034] Step S200: Obtain a multi-party information receiver and extract the feature vector representing the data granularity requirement in the multi-party information receiver.

[0035] Preferably, the multi-party information receiving end includes multiple terminal devices that need to receive processed data, which can be different types of data receiving devices or systems, such as hospital management systems (used to display patients' health data, diagnostic results, etc.), patient-side devices (patients' mobile phone applications), health monitoring devices (such as blood glucose meters, heart rate monitors, etc.), family doctors, mobile devices, etc. These different receiving ends (or terminals) have different data requirements, some require low-granularity information (such as trend charts), and some require high-granularity detailed data (such as detailed blood glucose change curves, image analysis results, etc.); then extract the feature vector representing the data granularity requirements from the multi-party information receiving end, wherein the feature vector is used to capture certain features of the information and represent the granularity requirements of each receiving end. Specifically, the granularity requirements refer to the requirements of different receiving ends for the level of data details. For example, the patient-side device may only need to provide brief information on health trends, while the hospital management system may require detailed diagnostic data.

[0036] Preferably, a feature vector is extracted from each receiving end to represent the data granularity requirements of the receiving end. By extracting the feature vector of the receiving end, the data granularity requirements of each receiving end are quantified. The information contained in the feature vector may include the data granularity requirements (for example, whether detailed data at each time point is required, or only the trend after aggregation is required), the data format (for example, whether charts, tables, or raw data are required), the complexity of the data (for example, whether deeply analyzed data is required, or only surface data), and the modal preference of the data. According to the specific needs of each receiving end, data of different granularities are generated and provided, so as to achieve more personalized and accurate health data transmission and display.

[0037] Step S300, clustering the multi-party information receiving terminals according to the feature vectors, outputting clustering results, configuring the number of the multiple transmission channels according to the number of classes of the clustering results, and establishing connections between the multiple transmission channels and the multi-party information receiving terminals.

[0038] Preferably, the receiving ends are grouped according to the eigenvectors through a clustering algorithm (such as K-means, hierarchical clustering, DBSCAN, etc.), that is, according to the similarity of the receiving end eigenvectors, the receiving ends with similar requirements are grouped into the same class. Specifically, K-means is used to cluster the multi-party information receiving ends. First, the number of clusters is determined. For example, it is preset to divide the receiving ends into 2 classes (including the receiving ends with low-granularity data requirements and the receiving ends with high-granularity data requirements). Then, K centroids are initialized to represent the centers of each class. In the first round of iteration, K-means randomly selects K data points as the initial centroids, and then assigns each receiving end to the closest centroid. Usually, the Euclidean distance is used to calculate the distance between each receiving end and the centroid, and it is assigned to the closest centroid. All receiving ends are assigned to a certain cluster, and K-means recalculates the new centroid of each cluster (that is, the average position of all data points assigned to this class), and then repeatedly executes the assignment of receiving ends to the closest centroid and the update of the centroid until the centroid position no longer changes or changes very little. Finally, the clustering result is output. For example, the receiving ends that require high-granularity data (such as hospital management systems) and the receiving ends that require low-granularity data (such as patient mobile applications), each cluster represents a class of receiving ends with similar requirements, and then different transmission channels are configured according to these clustering results.

[0039] Preferably, according to the number of classes in the clustering result, the corresponding number of transmission channels are configured. Each cluster corresponds to a transmission channel, that is, each receiving end with similar data requirements will receive data through an independent transmission channel. Among them, the transmission channel is used to transmit data with different granularities and formats according to the requirements of each class, and then based on the number of transmission channels, it is connected to the multi-party information receiving ends. Specifically, according to the class where each receiving end is located, these receiving ends are connected to the corresponding transmission channels. Each receiving end will receive the template data with the corresponding granularity through the transmission channel it is connected to according to its own data granularity requirements, ensuring that each receiving end receives the correct granularity data that meets its requirements, providing customized and efficient data transmission for each class of receiving ends, thereby effectively managing the requirements of different receiving ends and ensuring the efficiency and accuracy of data transmission.

[0040] Further, step S300 further includes step S310, where the eigenvector of each information receiving end includes data fineness requirements, data interaction complexity, data modality preference, and granularity level requirements; step S320, calculating the similarity between the eigenvectors of every two information receiving ends among the multi-party information receiving ends to obtain a similarity set; step S330, clustering the multi-party information receiving ends by using the similarity set and outputting a clustering result.

[0041] Preferably, the feature vector of each information receiving end includes the following four dimensions, namely the data fineness requirement, the data interaction complexity, the data modality preference, and the granularity level requirement. Specifically, the data fineness requirement refers to the requirement of the receiving end for the detail degree of data. For example, some receiving ends need fine-grained real-time data, while other receiving ends only need rough trend information; the data interaction complexity refers to the requirement of the receiving end for the complexity of data interaction. For example, some receiving ends need to perform complex interaction operations with data (such as visualization and in-depth analysis), while other receiving ends may only need basic data display; the data modality preference refers to the preference of the receiving end for different data modalities. For example, whether it prefers to process image data, sensor data, text data, etc.; the granularity level requirement refers to the requirement of the receiving end for the granularity level of data. For example, whether it needs raw data (low granularity) or aggregated data (high granularity).

[0042] Preferably, the similarity between the feature vectors of each information receiving end is calculated, which measures the similarity degree of the receiving ends in terms of data requirements. For example, the cosine similarity is used to calculate the similarity of two feature vectors. The closer the value is to 1, the more similar it is, and the value of 0 means completely dissimilar; or the Euclidean distance is used to calculate the distance between two feature vectors in a multi-dimensional space. The smaller the distance, the higher the similarity between the receiving ends. Then, the similarity values of multiple receiving ends are obtained, forming a similarity set; then, based on the similarity, the receiving ends are divided into K classes, where K is a preset clustering number. K-means clustering assigns the receiving ends with high similarity to the same class, that is, the feature vector of each receiving end will be assigned to a cluster according to their similarity. The receiving ends with higher similarity will be assigned to the same class so that the requirements of these receiving ends (such as data fineness, interaction complexity, etc.) can be processed similarly. Finally, the clustering result is output, including the class to which each receiving end belongs and the representative features of each class, such as the center (i.e., the centroid) of the class, so as to provide customized data processing and transmission solutions for different receiving ends.

[0043] Step S400, set the multiple data granularities of the multiple transmission channels, perform template processing according to the multiple data granularities respectively, output multiple granularity templates, download the multiple granularity templates to the multiple transmission channels, and perform templatization processing on the chronic disease resource data transmitted by the multiple transmission channels according to the multiple granularity templates, and transmit them to the multiple information receiving ends correspondingly.

[0044] Preferably, multiple data granularities are set for multiple transmission channels, that is, corresponding data granularities (including data detail levels or processing levels) are set for each transmission channel. Different receiving ends may have different requirements for data granularity. Data granularity can be divided into multiple levels. Low granularity, such as a health trend graph or summary information, is suitable for display to patients or users. Medium granularity, such as partial detailed information or periodic analysis data of health data, is suitable for use by family doctors, nurses, etc. High granularity, such as detailed physiological parameter time series data, imaging data, etc., is suitable for in-depth analysis by hospital management systems. Then, template processing is performed according to multiple data granularities, that is, corresponding data templates are generated for each granularity according to different data granularity requirements, and then multiple granularity templates are output. Each template represents a data format at a different level of detail (that is, each transmission channel will correspond to a granularity template), which may include a low granularity template (for displaying summary data, overview graphs, simplified trends, etc.), a medium granularity template (displaying periodic health data, partial detailed analysis charts), and a high granularity template (displaying detailed time series data, complete health records, medical images, etc.). Furthermore, data of different granularities is converted into a standard format suitable for display or processing to ensure that the display format and detail level of the data meet the requirements of the receiving end.

[0045] Preferably, the corresponding granularity template is downloaded for each transmission channel to ensure that data is transmitted in a predefined format. For example, the low granularity template is downloaded to transmission channel 1 for receiving ends that require low granularity data (such as patient applications); the medium granularity template is downloaded to transmission channel 2 for receiving ends that require medium granularity data (such as family doctors); the high granularity template is downloaded to transmission channel 3 for receiving ends that require high granularity data (such as hospital management systems); ensuring that each transmission channel can transmit appropriate templated data according to the requirements of the receiving end.

[0046] Preferably, when the platform transmits chronic disease resource data through a transmission channel, the templating process formats the original chronic disease resource data to meet the requirements and standards of the receiving end, that is, the original chronic disease data (such as a patient's physiological data, medical images, health monitoring data, etc.) is sorted and displayed according to the definition of the template, and finally transmitted to multiple information receiving ends through the corresponding transmission channel. For example, a patient application device (receiving low granularity data and viewing health trend graphs, etc.), a family doctor (receiving medium granularity data, such as analysis charts, periodic data, etc.), and a hospital management system (receiving high granularity data, such as detailed physiological data, medical images, the patient's complete health record, etc.), ensuring that the data received by different receiving ends meets the granularity and format required by them, thereby improving the data transmission and display effects in chronic disease resource management and meeting the needs of different users.

[0047] Further, step S400 further includes step S410, where the cloud processor performs templatization processing on the cross-modal database to generate an initialization template, and the initialization template includes the number of initialization template partitions, the initialization information hierarchy structure, the initialization data update frequency, and the initialization data display accuracy; step S420, adjusts the template parameters of the initialization template according to the multiple data granularities, and outputs multiple granularity templates, where the template parameters include the number of template partitions, the information hierarchy structure data, the update frequency, and the data display accuracy.

[0048] Preferably, the cloud processor performs templatization processing on the data in the cross-modal database (such as medical images, electronic health records, genomic data, sensor data, etc.) to generate an initialization template, that is, converts it into a standardized format and structure, ensuring that different types and formats of data can be uniformly presented to different users (such as patients, doctors, researchers, etc.), and solving the differences in data format, granularity, level of detail, etc. Specifically, the number of initialization template partitions, the initialization information hierarchy structure, the initialization data update frequency, and the initialization data display accuracy are defined. Among them, the number of initialization template partitions determines the partitioning and display areas of the data in the template. Each partition represents a different type of data display area, reflecting the level and complexity of data display. The more template partitions there are, the more refined the data display. For example, one partition is used to display the patient's basic information (such as name, age, gender, etc.), and another partition is used to display the health status (such as blood sugar level, heart rate, body temperature, etc.).

[0049] Preferably, the initialized information hierarchy defines the hierarchical structure of data in the template, that is, the display order of data from overview to details. For example, the top layer displays high-level health overview data (overall health status, trends, etc.); the secondary layer displays more detailed data (such as the results of the most recent examination, various physiological indicators of the patient, etc.); the bottom layer displays the most detailed data (such as daily health monitoring data, laboratory test data, etc.). The initialized data update frequency defines the frequency of data update, and different types of data have different requirements for update frequency. For example, for real-time monitoring data (such as the patient's blood sugar level, heart rate, etc.), it needs to be updated frequently, perhaps every minute or every hour; for static data (such as diagnostic reports, historical medical records, etc.), the update frequency is lower, perhaps once a week or once a month, to ensure the timeliness of the data and set reasonable update times according to different data types. The initialized data display precision defines the precision level of data display, that is, the degree of detail of the displayed data. For example, for the receiving end that requires detailed data display (such as doctors or researchers), it may display complete experimental results, time series data, etc.; for the receiving end that requires simplified display (such as the patient end or health monitoring device), it may only display trend charts or summary data; the data display precision can be flexibly adjusted according to the needs of the receiving end, so that the data display will not be too complex.

[0050] Preferably, different receiving ends have different requirements for data granularity, and then adjust each parameter of the template according to different data granularity requirements. Specifically, if the receiving end needs to display more data levels (such as detailed health analysis), it may increase the number of partitions of the template and divide the data into more display areas; for the receiving end that requires more detailed data (such as the hospital system), the hierarchical structure will be more complex and in-depth, and for the receiving end with lower granularity requirements (such as the patient application), the hierarchical structure may be simplified; for the receiving end of high-granularity data (such as the real-time monitoring system), the data update frequency is higher, while for low-granularity data (such as the health trend chart), the update frequency is lower; the receiving end of high-granularity data may need to display accurate time series data or detailed laboratory results, and its display precision will be higher, while the receiving end of low-granularity data may only need rough trends or overview data, and the display precision is set lower; finally, multiple granularity templates are generated, each template corresponding to a different requirement level, and different granularity data will be displayed according to the specific needs of the receiving end, thus ensuring that the data in the cross-modal database can provide the most suitable display format according to the needs of different receiving ends, while optimizing the efficiency of data processing and transmission.

[0051] Further, step S420 further includes step S421, where the multi-party information receiver performs a quality assessment on the received templatized chronic disease resource data to obtain a plurality of quality assessment indicators corresponding to the multi-party information receiver; step S422, feeds back the plurality of quality assessment indicators to the cloud processor, and updates the template parameters of the plurality of granular templates according to the cloud processor until the preset quality assessment indicators are met, and outputs the updated plurality of granular templates.

[0052] Preferably, the multi-party information receiver is used to perform a quality assessment on the templatized chronic disease resource data, that is, to check all aspects of the data to ensure the accuracy, integrity, timeliness, and availability of the data. This may include assessing whether the data is true and error-free and reflects the actual health status; whether the data is complete and lacks key parts; whether the data is consistent with other data sources and there are conflicts; whether the data is updated in a timely manner and reflects the latest health status; whether the displayed data meets the preset display accuracy requirements (such as accurate to days, hours, minutes, etc.); the quality assessment of the data by each receiver generates a plurality of quality assessment indicators, reflecting the quality status of the data in different dimensions. For example, for the patient-side application, more attention is paid to data accuracy and display accuracy, and for the hospital management system, more attention may be paid to data integrity, timeliness, and consistency.

[0053] Preferably, each receiver feeds back the assessment results of the data quality (i.e., a plurality of quality assessment indicators) to the cloud processor to help understand whether the current data quality meets the needs of different receivers and whether the template needs to be adjusted to improve the data quality. Then, the template parameters are adjusted according to the received quality assessment indicators to meet the quality requirements of the receiver. Specifically, if the data display is not clear enough, the display area in the template may be increased, or the size of the partition may be adjusted; if the data hierarchy is not clear enough, or more detailed information needs to be displayed, the cloud processor will adjust the hierarchical structure of the template to make the data layering more reasonable; if the feedback shows that the timeliness of data update is insufficient, the cloud processor may increase the frequency of data update to ensure that the receiver receives the latest data; if the data display accuracy is insufficient, the cloud processor can adjust the display accuracy of the template, such as providing a more detailed timestamp, a more accurate numerical display, etc.

[0054] Preferably, the cloud processor continuously optimizes the template according to the feedback from the receiving end until the data quality meets the preset quality assessment criteria. The preset quality assessment indicators are the set ideal data quality standards, which may include that the accuracy rate of the data reaches a certain standard, the timeliness requirements of the data, such as being updated every hour, the integrity requirements of the data, such as no missing data fields, and whether the display accuracy meets the expectations, such as being accurate to the hour or minute. When all template parameters are adjusted to meet the preset quality standards, the cloud processor will generate and output multiple updated granular templates and redistribute them to the corresponding receiving ends to ensure the best data quality and ensure that different receiving ends can receive data displays that meet their quality requirements.

[0055] Further, step S422 further includes step A: using the multi-party information receiving end as an edge node and the cloud processor as a central node to construct a resource topology management platform; step B: performing update frequency synchronization processing on the granular templates of the multiple transmission channels based on the resource topology management platform.

[0056] Preferably, use multi-party information receiving ends (such as hospital management systems, patients' mobile applications, health monitoring devices, etc.) as edge nodes, which are responsible for receiving and displaying data according to their own needs (such as different data granularities or data formats); use the cloud processor as a central node, which is responsible for centrally managing tasks such as data storage, processing, template generation, and update, including processing the aggregation, templatization of cross-modal data, and adjusting the granularity, accuracy, update frequency, etc. of data transmission according to the feedback from the receiving end; construct a resource topology management platform, which is responsible for monitoring and coordinating the data flow, template management, and update between the edge nodes (receiving ends) and the central node (cloud processor). Specifically, it includes node management, which manages the resources of each receiving end (edge node) and the cloud processor (central node), including storage capacity, computing power, network connection, etc.; data flow management, which coordinates the data transmission from the cloud to the edge nodes to ensure the accuracy, timeliness, and security of the data; template management, which manages the generation, update, and distribution of cross-modal data templates to ensure that different receiving ends can obtain data display templates suitable for their own needs), thereby more efficiently coordinating and managing the transmission, processing, and update of data.

[0057] Preferably, different transmission channels are responsible for transmitting data templates of different granularities (such as low granularity, medium granularity, and high granularity). Each transmission channel will transmit the data template suitable for a specific receiving end to the target receiving end, and then the resource topology management platform performs synchronization processing on the update frequencies of the granularity templates of multiple transmission channels. Specifically, due to different data granularities and the different data requirements of receiving ends, the data update frequencies may vary. For example, data for real-time monitoring (such as blood sugar level, heart rate, etc.) may need to be updated once a minute, while summary data (such as health trends) may only need to be updated daily or weekly. Through the coordination of the platform, the templates of different transmission channels (low granularity, medium granularity, high granularity) are updated at a consistent frequency, ensuring that each receiving end can obtain the latest template data that meets its needs within an appropriate time, improving the efficiency, timeliness, and reliability of data transmission.

[0058] In the foregoing, reference is made to Figure 1 which describes in detail the cross-modal data generation-driven chronic disease management method according to an embodiment of the present invention. Next, reference will be made to Figure 2 describe the cross-modal data generation-driven chronic disease management platform according to an embodiment of the present invention.

[0059] The cross-modal data generation-driven chronic disease management platform according to an embodiment of the present invention is used to solve the technical problems in the prior art, such as inconsistent cross-modal data granularity and modal semantics, and difficulty in flexibly generating appropriate granularity templates according to requirements, resulting in low information transmission efficiency and poor accuracy and timeliness of chronic disease management, and achieves the technical effect of improving data transmission efficiency and the accuracy and timeliness of chronic disease management. As Figure 2 shown, the cross-modal data generation-driven chronic disease management platform includes: a cross-modal database access unit 10, a feature vector extraction unit 20, a transmission channel configuration unit 30, and a granularity template output unit 40.

[0060] The cross-modal database access unit 10 is used to access the cross-modal database, which is stored in a cloud processor. The cloud processor includes multiple transmission channels; the feature vector extraction unit 20 is used to obtain multiple information receivers and extract the feature vectors representing the data granularity requirements in the multiple information receivers; the transmission channel configuration unit 30 is used to cluster the multiple information receivers according to the feature vectors, output a clustering result, configure the number of the multiple transmission channels according to the number of classes in the clustering result, and establish connections between the multiple transmission channels and the multiple information receivers; the granularity template output unit 40 is used to set multiple data granularities for the multiple transmission channels, perform template processing according to the multiple data granularities respectively, output multiple granularity templates, download the multiple granularity templates to the multiple transmission channels, and perform templatization processing on the chronic disease resource data transmitted by the multiple transmission channels and transmit it to the multiple information receivers correspondingly.

[0061] Next, the specific configuration of the cross-modal database access unit 10 will be described in detail. The cross-modal database access unit 10 further includes: performing high-dimensional feature space mapping on the data of each modality in the cross-modal database, and outputting a multi-modal high-dimensional feature data set corresponding to multiple modalities; using a functional adversarial network to perform generative alignment on the multi-modal high-dimensional feature data set, and outputting an aligned multi-modal feature data set; using the aligned multi-modal feature data set to perform modal data feature fusion to update the cross-modal database.

[0062] Next, the specific configuration of the cross-modal database access unit 10 will be described in detail. The cross-modal database access unit 10 further includes: wherein, the functional adversarial network includes a generator and a discriminator, the output end of the generator is connected to the input end of the discriminator, and the output end of the discriminator is connected to the input end of the generator; performing adversarial training on the generator and the discriminator, obtaining the modal data generated by the generator, inputting the generated modal data into the discriminator, calculating the difference distribution from the sample modal data, and performing parameter learning with the goal of being less than a preset difference distribution to obtain a trained functional adversarial network.

[0063] Next, the specific configuration of the cross-modal database access unit 10 will be described in detail. The cross-modal database access unit 10 further includes: calculating the semantic difference degree between the features in each modality of the aligned multi-modal feature data set; storing the features with a semantic difference degree greater than or equal to the preset semantic difference degree in parallel, and storing the features with a semantic difference degree less than the preset semantic difference degree in a covering manner.

[0064] Next, the specific configuration of the granularity template output unit 40 will be described in detail. The granularity template output unit 40 further includes: the cloud processor performs templatization processing on the cross-modal database to generate an initialization template, and the initialization template includes the number of initialization template partitions, the initialization information hierarchy structure, the initialization data update frequency, and the initialization data display accuracy; adjusting the template parameters of the initialization template according to the multiple data granularities to output multiple granularity templates, where the template parameters include the number of template partitions, the information hierarchy structure data, the update frequency, and the data display accuracy.

[0065] Next, the specific configuration of the granularity template output unit 40 will be further described in detail. The granularity template output unit 40 further includes: using the multi-party information receiver as an edge node and the cloud processor as a central node to construct a resource topology management platform; performing update frequency synchronization processing on the granularity templates of the multiple transmission channels based on the resource topology management platform.

[0066] Next, the specific configuration of the granularity template output unit 40 will be further described in detail. The granularity template output unit 40 further includes: obtaining a warning matching policy logic library according to the hierarchical warning mechanism; performing labeled classification on the warning matching policy logic library according to the battery fault label set to obtain a battery warning labeled policy library; performing policy matching and parsing on the battery module fault prediction parameters based on the battery warning labeled policy library to determine a battery fault warning policy, and the battery fault warning policy includes a warning signal type and a warning method level.

[0067] Next, the specific configuration of the transmission channel configuration unit 30 will be further described in detail. The transmission channel configuration unit 30 further includes: where the feature vector of each information receiver includes data fineness requirements, data interaction complexity, data modality preference, and granularity level requirements; calculating the similarity between the feature vectors of every two information receivers among the multi-party information receivers to obtain a similarity set; clustering the multi-party information receivers using the similarity set to output a clustering result.

[0068] The cross-modal data generation-driven chronic disease management platform provided by the embodiments of the present invention can execute the cross-modal data generation-driven chronic disease management method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0069] Although the present application makes various references to certain modules in the platform according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The various units and modules included are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0070] The above specific embodiments do not constitute a limitation to the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to the design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A chronic disease management method driven by cross-modal data generation, characterized in that: The method comprises: Accessing a cross-modal database, wherein the cross-modal database is stored in a cloud processor, wherein the cloud processor includes a plurality of transmission channels; Acquire a multi-party information receiving end, and extract a feature vector representing a data granularity requirement from the multi-party information receiving end; Clustering the multi-party information receiving terminals according to the feature vectors, outputting clustering results, configuring the number of the multiple transmission channels according to the number of clustering results, and establishing connections between the multiple transmission channels and the multi-party information receiving terminals; Setting multiple data granularities of the multiple transmission channels, performing template processing respectively according to the multiple data granularities, outputting multiple granularity templates, downloading the multiple granularity templates to the multiple transmission channels, performing template processing on the chronic disease resource data transmitted by the multiple transmission channels according to the multiple granularity templates, and transmitting the data to the multiple information receiving terminals accordingly; After accessing the cross-modal database, the methods include: Performing high-dimensional feature space mapping on the data of each modality in the cross-modal database, and outputting a multimodal high-dimensional feature data set corresponding to multiple modalities; Generatively aligning the multimodal high-dimensional feature dataset using a functional adversarial network, and outputting an aligned multimodal feature dataset; Using the aligned multimodal feature data set to perform modal data feature fusion and update the cross-modal database; Methods for training functional-adversarial networks include: The functional adversarial network includes a generator and a discriminator, the output end of the generator is connected to the input end of the discriminator, and the output end of the discriminator is connected to the input end of the generator; The generator and the discriminator are subjected to adversarial training, the modal data generated by the generator is obtained, the generated modal data is input into the discriminator, the difference distribution with the sample modal data is calculated, parameter learning is performed with the goal of being less than a preset difference distribution, and a trained functional-adversarial network is obtained.

2. The cross-modal data generation-driven chronic disease management method according to claim 1, characterized in that: The method of performing modal data feature fusion using the aligned multimodal feature data set to update the cross-modal database includes: Calculate the semantic difference between the features in each modality of the aligned multimodal feature dataset; Features whose semantic difference is greater than or equal to a preset semantic difference are stored in parallel, and features whose semantic difference is less than the preset semantic difference are stored in an overwritten manner.

3. The cross-modal data generation-driven chronic disease management method according to claim 1, characterized in that: Performing template processing respectively according to the multiple data granularities and outputting multiple granularity templates, the method includes: The cloud processor performs template processing on the cross-modal database to generate an initialization template, wherein the initialization template includes the number of initialization template partitions, the initialization information hierarchical structure, the initialization data update frequency, and the initialization data display accuracy; The template parameters of the initialization template are adjusted according to the multiple data granularities to output multiple granularity templates, wherein the template parameters include the number of template partitions, information hierarchical structure data, update frequency, and data display accuracy.

4. The cross-modal data generation-driven chronic disease management method according to claim 3, characterized in that: After outputting the multiple granularity templates, the method further comprises: The multi-party information receiving end performs quality assessment on the received templated chronic disease resource data, and obtains a plurality of quality assessment indicators corresponding to the multi-party information receiving end; The multiple quality assessment indicators are fed back to the cloud processor, and the template parameters of the multiple granularity templates are updated according to the cloud processor until the preset quality assessment indicators are met, and the updated multiple granularity templates are output.

5. The cross-modal data generation-driven chronic disease management method according to claim 4, characterized in that: The multi-party information receiving end is used as an edge node, and the cloud processor is used as a central node to build a resource topology management platform; Based on the resource topology management platform, update frequency synchronization processing is performed on the granularity templates of the multiple transmission channels.

6. The cross-modal data generation-driven chronic disease management method according to claim 1, characterized in that: Clustering the multiple information receiving terminals according to the feature vectors and outputting the clustering results. include: Among them, the feature vector of each information receiving end includes data precision requirements, data interaction complexity, data modality preference, and granularity level requirements; Calculating the similarity between feature vectors of every two information receiving ends among the multiple information receiving ends to obtain a similarity set; The multiple information receiving terminals are clustered using the similarity set, and a clustering result is output.

7. A cross-modal data generation-driven chronic disease management platform, characterized by: The platform is used to implement the cross-modal data generation-driven chronic disease management method according to any one of claims 1 to 6, and the platform includes: A cross-modal database access unit, used to access a cross-modal database, wherein the cross-modal database is stored in a cloud processor, and the cloud processor includes a plurality of transmission channels; A feature vector extraction unit, used to obtain multiple information receiving terminals and extract feature vectors representing data granularity requirements from the multiple information receiving terminals; a transmission channel configuration unit, configured to cluster the multi-party information receiving terminals according to the feature vectors, output a clustering result, configure the number of the multiple transmission channels according to the number of classes of the clustering result, and establish connections between the multiple transmission channels and the multi-party information receiving terminals; A granularity template output unit is used to set multiple data granularities of the multiple transmission channels, perform template processing according to the multiple data granularities, output multiple granularity templates, download the multiple granularity templates to the multiple transmission channels, perform template processing on the chronic disease resource data transmitted by the multiple transmission channels according to the multiple granularity templates, and transmit them to the multiple information receiving ends accordingly.

Images