Multi-modal medical image intelligent filing method and system

Abstract

The invention relates to the technical field of medical information, and discloses a multi-modal medical image intelligent filing method and system, and the method comprises the steps: dynamically converting an original data file of medical image equipment into a standard three-dimensional matrix; extracting a multi-channel gradient feature map and carrying out organ segmentation, and carrying out contour anomaly correction on the multi-channel gradient feature map after organ segmentation based on anatomical topology constraints to obtain an anatomical structure map; fitting the time information and the shadow intensity into a metabolic curve of the patient, and encoding the metabolic curve into a metabolic activity vector of the patient; extracting pathological abnormality features, and fusing the pathological abnormality features and the metabolic activity vector to obtain a pathological label; constructing an extensible index tree of the patient; binding the standard three-dimensional matrix with the extensible index tree, encrypting fragments and storing the fragments to medical cloud nodes; according to the invention, the efficiency of multi-modal medical image intelligent filing can be improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a multimodal medical image intelligent archiving method and system. Background Art

[0002] In the field of medical information technology, multimodal medical imaging data is characterized by large data volumes, diverse formats, and complex modalities. Existing archiving methods lack a unified, standardized conversion mechanism for processing raw data generated by different devices, resulting in cumbersome data processing workflows and difficulty achieving efficient preprocessing. Furthermore, traditional methods are unable to effectively extract and fuse multidimensional information such as anatomical structures and metabolic activities contained in multimodal images. This makes it impossible to accurately reflect the inherent relationships of the data when constructing indexes and storing them, thus affecting archiving efficiency.

[0003] Furthermore, existing archiving systems, when faced with massive amounts of multimodal medical imaging data, rely on a relatively simple indexing approach and lack dynamic expansion and optimization mechanisms, making it difficult to quickly locate and retrieve target data. Furthermore, during data encryption and storage, there is a lack of targeted sharding strategies and integrity verification mechanisms. This not only increases the storage and transmission burden but can also hinder the archiving process due to inefficient data verification, further reducing the overall efficiency of intelligent archiving of multimodal medical images. Summary of the Invention

[0004] The present invention provides a multimodal medical image intelligent archiving method and system, the main purpose of which is to solve the problem of low efficiency in the intelligent archiving of multimodal medical images.

[0005] To achieve the above objectives, the present invention provides a multimodal medical image intelligent archiving method, comprising: S1. Dynamically convert the raw data files of medical imaging equipment into a standard three-dimensional matrix; S2. extracting a multi-channel gradient feature map from the standard three-dimensional matrix, performing organ segmentation on the multi-channel gradient feature map, and correcting contour anomalies on the multi-channel gradient feature map after organ segmentation based on anatomical topological constraints to obtain an anatomical structure atlas of the patient; S3. fitting the time information and shadow intensity of the dynamic image in the standard three-dimensional matrix into a metabolic curve of the patient, and encoding the metabolic curve into a metabolic activity vector of the patient; S4. extracting pathological abnormality features from the anatomical structure atlas, fusing the pathological abnormality features with the metabolic activity vector, and obtaining a pathological label for the patient; S5. Constructing an extensible index tree for the patient based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file; S6. Bind the standard three-dimensional matrix to the extensible index tree, and encrypt and store the fragments in the medical cloud node.

[0006] In a preferred embodiment, the dynamic conversion of the raw data file of the medical imaging device into a standard three-dimensional matrix includes: Parsing the header information of the original data file to obtain the spatial resolution parameter of the original data file; performing isotropic resampling on the image data in the original data file according to the spatial resolution parameter; The isotropically resampled image data is mapped to a standard coordinate system to generate a standard three-dimensional matrix of the original data file.

[0007] In a preferred embodiment, the performing organ segmentation on the multi-channel gradient feature map and correcting contour abnormalities on the multi-channel gradient feature map after organ segmentation based on anatomical topology constraints to obtain an anatomical structure atlas of the patient includes: Fusing shallow high-resolution features and deep semantic features of the multi-channel gradient feature map to obtain a patient's organ probability heat map; performing contour abnormality correction on the binary organ probability heat map based on anatomical topology constraints to obtain a binary organ mask atlas of the patient; The outer surface triangular mesh of the binary organ mask atlas is mapped back to the spatial coordinate system of the standard three-dimensional matrix to obtain the anatomical structure atlas of the patient.

[0008] In a preferred embodiment, fitting the time information and shadow intensity of the dynamic image in the standard three-dimensional matrix into the metabolic curve of the patient, and encoding the metabolic curve into the metabolic activity vector of the patient, comprises: Extract voxel intensity values ​​at consecutive time points in dynamic images; identifying the shadow intensity of the target organ region in the dynamic image by using the voxel intensity value; The time information and the shadow intensity are fitted into the metabolic curve of the patient, wherein the fitting calculation formula is as follows:

[0009] Where, for Plasma input at any time, for The contrast agent concentration in the target organ area is linearly converted from the shadow intensity at each moment. is the contrast agent transmembrane transport rate constant, is the tissue clearance rate constant, is the time information, is the differential symbol, is the metabolic profile of the patient; After the curvature features of the metabolic curve are spliced ​​into a high-dimensional vector, the dimension is reduced to the metabolic activity vector of the patient.

[0010] In a preferred embodiment, extracting the pathological abnormality features of the anatomical structure atlas, fusing the pathological abnormality features with the metabolic activity vector, and obtaining the pathological label of the patient includes: Locating the abnormal area in the anatomical structure atlas, and using the morphological indexes and texture indexes of the abnormal area as primary pathological features; Mapping the metabolic activity vector into a metabolic pathology feature vector through a fully connected layer; The primary pathological feature is fused with the metabolic pathological feature vector to obtain the pathological label of the patient.

[0011] In a preferred embodiment, the constructing of the patient's extensible index tree based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file includes: Generate a first level with the pathology label as the root node; Encode the device type as an orthogonal feature vector at the second level; Generate the third-level leaf nodes based on the frequency domain features of the imaging modality; An extensible index tree of the patient is constructed using the root node, the orthogonal feature vectors, and the leaf nodes.

[0012] In a preferred embodiment, after constructing the patient's extensible index tree based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file, the following steps are included: Pruning and merging the leaf nodes whose similarity is higher than the redundancy threshold; The node weights of the scalable index tree are dynamically updated based on data timeliness.

[0013] In a preferred embodiment, the encrypted fragments are stored in the medical cloud node, including: encrypting the bound extensible index tree using the patient's biometric key; Dividing the bound extensible index tree into blocks according to the organ regions of the standard three-dimensional matrix to obtain data blocks of the original data file; Distribute and store data blocks to different medical cloud nodes.

[0014] In a preferred embodiment, the binding of the standard three-dimensional matrix to the scalable index tree and encrypted fragment storage to the medical cloud node includes: When the original data file is archived, restoring the data block from the medical cloud node; Reconstructing the data block encryption matrix to obtain a hash value of the archived data block; Comparing the deviation between the hash value and the original hash value of the data block to obtain the deviation value of the archived data block; When the deviation value is lower than the data damage threshold, a successful archiving signal is triggered.

[0015] In order to solve the above problems, the present invention further provides a multimodal medical image intelligent archiving system, the system comprising: Image conversion module, used to dynamically convert raw data files of medical imaging equipment into standard three-dimensional matrices; an anatomical structure atlas generation module, configured to extract a multi-channel gradient feature map from the standard three-dimensional matrix, perform organ segmentation on the multi-channel gradient feature map, and perform contour abnormality correction on the multi-channel gradient feature map after organ segmentation based on anatomical topological constraints to obtain an anatomical structure atlas of the patient; a metabolic activity vector generation module, configured to fit the time information and shadow intensity of the dynamic image in the standard three-dimensional matrix into a metabolic curve of the patient, and encode the metabolic curve into a metabolic activity vector of the patient; a pathology label generation module, configured to extract pathology abnormality features from the anatomical structure atlas, fuse the pathology abnormality features with the metabolic activity vector, and obtain a pathology label for the patient; an extensible index tree construction module, configured to construct an extensible index tree for the patient based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file; The data archiving module is used to bind the standard three-dimensional matrix with the extensible index tree and store the encrypted fragments in the medical cloud node.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention achieves standardized preprocessing of multimodal imaging data by dynamically converting raw data into a standard three-dimensional matrix, avoiding time-consuming processing due to differences in data formats. It also improves the consistency and efficiency of data processing through isotropic resampling and standard coordinate system mapping. In the feature extraction and analysis phase, multi-channel gradient features are integrated with anatomical topological constraints for organ segmentation and contour correction. Metabolic curves are fitted using temporal information and shadow intensity and encoded as metabolic activity vectors. This allows for efficient extraction of multidimensional pathological features, laying the foundation for the subsequent rapid generation of pathological labels and indexing, effectively reducing the time cost of feature analysis.

[0017] 2. The present invention constructs an extensible index tree based on pathology labels, device types, and imaging modalities, and optimizes the index structure through pruning, merging, and dynamic weight updates, significantly improving data retrieval efficiency. It adopts biometric key encryption and fragmented storage by organ region to achieve distributed and efficient data storage while ensuring data security. It also verifies archiving integrity through hash value comparison, ensuring a fast and accurate archiving process, significantly improving the overall efficiency of intelligent archiving of multimodal medical images. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of a process for intelligent archiving of multimodal medical images according to an embodiment of the present invention; Figure 2 A functional module diagram of a multimodal medical image intelligent archiving system provided by one embodiment of the present invention; The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0019] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] The embodiment of the present application provides a multimodal medical image intelligent archiving method. The execution subject of the multimodal medical image intelligent archiving method includes but is not limited to at least one of the electronic devices such as the server and the terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the multimodal medical image intelligent archiving method can be executed by software or hardware installed on the terminal device or the server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0021] Reference Figure 1 FIG. 1 is a flow chart of a multimodal medical image intelligent archiving method according to an embodiment of the present invention. In this embodiment, the multimodal medical image intelligent archiving method includes: S1. Dynamically convert the raw data files of medical imaging equipment into a standard three-dimensional matrix; In an embodiment of the present invention, the step of dynamically converting the raw data file of the medical imaging device into a standard three-dimensional matrix includes: Parsing the header information of the original data file to obtain the spatial resolution parameter of the original data file; performing isotropic resampling on the image data in the original data file according to the spatial resolution parameter; The isotropically resampled image data is mapped to a standard coordinate system to generate a standard three-dimensional matrix of the original data file.

[0022] Specifically, the original data file is opened, and the header information is read according to the fixed position and rules specified by the file format, and the content describing the spatial resolution is accurately extracted from the header information, thereby obtaining the spatial resolution parameters of the original data file.

[0023] Furthermore, based on the spatial resolution parameters of the obtained original data file, the image data in the original data file is isotropically resampled, that is, the image data is resampled in the same proportion and manner in the horizontal and vertical directions, so that the resolution of the image in all directions is consistent.

[0024] Furthermore, the isotropically resampled image data is mapped to the corresponding position of the standard coordinate system according to the definition and transformation rules of the standard coordinate system. Through this mapping method, the standard three-dimensional matrix of the original data file is finally constructed and generated.

[0025] In general, the raw data files of medical imaging equipment are dynamically converted into a standard three-dimensional matrix. The spatial resolution parameters can be obtained by parsing the head information, and the image data is isotropically resampled based on the parameters and mapped to the standard coordinate system to generate a three-dimensional matrix in a unified format.

[0026] In general, this process eliminates the format differences and spatial resolution inconsistencies of raw data from different devices, enables multimodal imaging data to have a standardized storage structure and spatial reference system, avoids preprocessing process redundancy and compatibility issues caused by inconsistent data formats, greatly reduces the time consumption of data conversion, and provides a standardized and unified data foundation for subsequent organ segmentation, feature extraction and other operations, thereby effectively improving the overall processing efficiency of multimodal medical image archiving.

[0027] In general, after the original data is unified into a standard three-dimensional matrix through standardized transformation, it can be directly bound to the subsequently constructed extensible index tree to achieve efficient association between data and index.

[0028] In general, this standardized processing enables imaging data of different modalities and different devices to be stored and managed in a consistent manner. When encrypted and fragmented data is stored in medical cloud nodes, it can be efficiently divided and distributed based on the structural characteristics of the standard matrix, reducing the format conversion overhead during data storage and transmission, and further improving the efficiency of the archiving process and the convenience of data management.

[0029] S2. extracting a multi-channel gradient feature map from the standard three-dimensional matrix, performing organ segmentation on the multi-channel gradient feature map, and correcting contour anomalies on the multi-channel gradient feature map after organ segmentation based on anatomical topological constraints to obtain an anatomical structure atlas of the patient; In an embodiment of the present invention, the step of performing organ segmentation on the multi-channel gradient feature map and correcting contour abnormalities on the multi-channel gradient feature map after organ segmentation based on anatomical topology constraints to obtain an anatomical structure atlas of the patient includes: Fusing shallow high-resolution features and deep semantic features of the multi-channel gradient feature map to obtain a patient's organ probability heat map; performing contour abnormality correction on the binary organ probability heat map based on anatomical topology constraints to obtain a binary organ mask atlas of the patient; The outer surface triangular mesh of the binary organ mask atlas is mapped back to the spatial coordinate system of the standard three-dimensional matrix to obtain the anatomical structure atlas of the patient.

[0030] Specifically, the shallow high-resolution features and deep semantic features in the multi-channel gradient feature map are integrated, and the detail information contained in the shallow high-resolution features and the semantic information contained in the deep semantic features are complemented and combined to obtain an organ probability heat map that can reflect the probability distribution of the patient's organs.

[0031] Furthermore, the binarized organ probability heat map is processed according to the anatomical topological constraints to check whether the contour of the organ probability heat map conforms to the normal topological relationship of the human anatomical structure. For the abnormal contour part that does not conform to the normal anatomical topological relationship, correction and adjustment are performed to obtain an accurate binary organ mask map of the patient.

[0032] Furthermore, the outer surface of the binary organ mask atlas is constructed into a triangular mesh form, and then each point on the triangular mesh is converted to the corresponding position in the spatial coordinate system of the standard three-dimensional matrix according to the rules and mapping relationship of the spatial coordinate system of the standard three-dimensional matrix, and finally an anatomical structure atlas that can present the patient's anatomical structure is generated.

[0033] In general, by extracting multi-channel gradient feature maps from a standard three-dimensional matrix and performing organ segmentation and contour correction, accurate anatomical structure maps can be obtained efficiently.

[0034] In general, this process generates organ probability heat maps by fusing shallow high-resolution features with deep semantic features, and corrects contour abnormalities in combination with anatomical topological constraints, avoiding the problem of missegmentation caused by noise or feature loss in traditional segmentation, reducing the time spent on manual intervention or repeated processing, and making the organ segmentation results more consistent with anatomical logic, thereby improving processing efficiency in the feature extraction stage.

[0035] In general, mapping the outer surface triangular mesh of the binary organ mask atlas back to the spatial coordinate system generates a structured anatomical structure atlas, which provides an accurate spatial positioning basis for subsequent pathological abnormality feature extraction, avoids feature extraction deviation caused by structural ambiguity, and shortens the time cost of pathological analysis.

[0036] In general, the anatomical structure atlas generated by the above operations has a standardized spatial coordinate system and precise organ contours, which can be directly and quickly integrated with multidimensional data such as metabolic activity vectors to accelerate the generation process of pathological labels.

[0037] In general, structured graph information facilitates the subsequent construction of an extensible index tree, using more accurate organ structures as the basis for index nodes, thereby improving the efficiency of index construction and retrieval accuracy.

[0038] In general, ensuring the rationality of the contour through anatomical topological constraints can also reduce data verification errors caused by structural abnormalities during the archiving process, further optimize the smoothness of the archiving process, and improve the archiving efficiency of multimodal medical images from feature extraction to index construction.

[0039] S3. fitting the time information and shadow intensity of the dynamic image in the standard three-dimensional matrix into a metabolic curve of the patient, and encoding the metabolic curve into a metabolic activity vector of the patient; In an embodiment of the present invention, fitting the time information and shadow intensity of the dynamic image in the standard three-dimensional matrix into the metabolic curve of the patient, and encoding the metabolic curve into the metabolic activity vector of the patient, includes: Extract voxel intensity values ​​at consecutive time points in dynamic images; identifying the shadow intensity of the target organ region in the dynamic image by using the voxel intensity value; The time information and the shadow intensity are fitted into the metabolic curve of the patient, wherein the fitting calculation formula is as follows:

[0040] Where, for Plasma input at any time, for The contrast agent concentration in the target organ area is linearly converted from the shadow intensity at each moment. is the contrast agent transmembrane transport rate constant, is the tissue clearance rate constant, is the time information, is the differential symbol, is the metabolic profile of the patient; After the curvature features of the metabolic curve are spliced ​​into a high-dimensional vector, the dimension is reduced to the metabolic activity vector of the patient.

[0041] Specifically, in dynamic images, the intensity values ​​of each voxel at consecutive time points are obtained in chronological order. These intensity values ​​represent the signal intensity of the corresponding position at different times.

[0042] Furthermore, based on the extracted voxel intensity values, the target organ area is found in the dynamic image, and statistical analysis is performed on all voxel intensity values ​​in the target organ area to determine the shadow intensity in the area. This shadow intensity can reflect the signal characteristics of the target organ in the image.

[0043] Furthermore, the time information corresponding to each time point is associated with the shadow intensity of the target organ area at that time point. Time is used as the horizontal axis and the shadow intensity is used as the vertical axis. These data points are connected to form a metabolic curve that can reflect the changes in the patient's target organ metabolism over time.

[0044] Furthermore, the curvature of the metabolic curve, that is, the curvature feature, is observed, and the curvature features at different positions are arranged in sequence and spliced ​​into a high-dimensional vector containing a lot of information. Then, a specific method is used to remove redundant information in the high-dimensional vector and reduce its dimension to obtain a streamlined metabolic activity vector that can represent the patient's metabolic activity.

[0045] Specifically, in the calculation formula of the fitting, the time information in the formula Directly obtain from the continuous time points recorded in the dynamic image, Plasma infusion at any moment It is obtained by measuring the amount of plasma entering the blood circulation system at each corresponding time point.

[0046] Further, Contrast agent concentration in the target organ area linearly converted from shadow intensity at any moment , which first obtains the shadow intensity of the target organ area in the dynamic image, and then converts the shadow intensity into contrast agent concentration according to a pre-set fixed linear conversion rule.

[0047] Furthermore, the contrast agent transmembrane transport rate constant and tissue clearance rate constant By experimentally measuring a large number of known data samples, the numerical value that can most accurately describe the contrast agent transmission and clearance process in the target organ area is determined.

[0048] Furthermore, this formula describes how the contrast agent concentration in the target organ region of the patient changes with time.

[0049] Further, Indicates At that moment, the amount of contrast agent brought by plasma infusion into the target organ area, Indicates The amount of contrast agent removed by the tissue in the target organ area at the time.

[0050] Further, The metabolic curve of the patient represented by the figure reflects the rate of change of the contrast agent concentration in the target organ area over time, that is, the time information and shadow intensity are fitted through such a relationship to show the metabolic process of the patient's target organ.

[0051] Furthermore, as time goes by, Greater than When the contrast agent is injected into the plasma, it means that the amount of contrast agent entering the body is greater than the amount of contrast agent cleared from the tissue. When the value of is positive, the contrast agent concentration in the target organ area will increase with time; Less than When tissue clearance is greater than input, the patient's metabolic curve If the value of is negative, the contrast agent concentration in the target organ area will decrease over time; equal When the two are balanced, the patient's metabolic curve The contrast agent concentration in the target organ area remains unchanged.

[0052] In general, by extracting voxel intensity values ​​at continuous time points in dynamic images and fitting metabolic curves, the metabolic dynamic characteristics of the patient's organs can be efficiently captured.

[0053] In general, this process uses the kinetic model formula to perform parameterized fitting of time information and shadow intensity, avoiding the complex point-by-point analysis of time series data in traditional methods and significantly shortening the computational time for metabolic feature extraction.

[0054] In general, the curvature features of the metabolic curve are spliced ​​and reduced to metabolic activity vectors, and the metabolic information is compressed and stored in the form of structured vectors, which reduces data redundancy and enables metabolic features to participate in subsequent pathological analysis in a more compact and efficient manner, laying the foundation for the rapid generation of pathological labels.

[0055] In general, the encoding method of metabolic activity vectors can be directly integrated with the pathological features of anatomical structure maps to accelerate the generation efficiency of pathological labels.

[0056] In general, this structured feature representation makes it easier to incorporate metabolic information as an independent dimension into the index system when subsequently constructing an extensible index tree, thereby improving the feature expression capabilities of the index nodes.

[0057] In general, the metabolic activity vector data volume after dimensionality reduction is small, which can reduce data transmission and storage overhead when stored in encrypted shards. Combined with the structural characteristics of the standardized three-dimensional matrix, efficient binding and distributed management of metabolic characteristics and imaging data can be achieved, and the archiving efficiency of multimodal medical images can be optimized from feature extraction to data storage.

[0058] S4. extracting pathological abnormality features from the anatomical structure atlas, fusing the pathological abnormality features with the metabolic activity vector, and obtaining a pathological label for the patient; In an embodiment of the present invention, extracting the pathological abnormality features of the anatomical structure atlas, fusing the pathological abnormality features with the metabolic activity vector, and obtaining the pathological label of the patient includes: Locating the abnormal area in the anatomical structure atlas, and using the morphological indexes and texture indexes of the abnormal area as primary pathological features; Mapping the metabolic activity vector into a metabolic pathology feature vector through a fully connected layer; The primary pathological feature is fused with the metabolic pathological feature vector to obtain the pathological label of the patient.

[0059] Specifically, a comprehensive and detailed observation and analysis of the anatomical structure atlas is conducted, and based on the morphology and characteristic standards of the normal human anatomical structure, areas in the atlas that do not conform to the normal structure are found and identified as abnormal areas.

[0060] Furthermore, for each abnormal area, its morphological indicators such as size and shape are measured, and its texture indicators such as grayscale distribution and texture direction are analyzed. These morphological indicators and texture indicators are integrated as primary pathological features.

[0061] Furthermore, the patient's metabolic activity vector obtained previously is input into the fully connected layer. Each node in the fully connected layer is connected to all elements in the metabolic activity vector. The metabolic activity vector is processed and transformed through the pre-set connection weights and activation methods between nodes, and the information in the metabolic activity vector is recombined and refined, and finally the metabolic pathology feature vector is output.

[0062] Furthermore, the acquired primary pathological features and metabolic pathological feature vectors are merged, and the anatomical structure information contained in the primary pathological features and the metabolic activity information reflected by the metabolic pathological feature vectors are fused. Through specific combination rules, this information is integrated into a whole, thereby obtaining a pathological label that can comprehensively reflect the patient's pathological condition.

[0063] In general, by locating abnormal areas in the anatomical structure atlas and using morphological and texture indicators as primary pathological features, we can directly focus on the key pathological areas, avoid redundant calculations of full-image feature extraction, and significantly shorten the feature extraction time.

[0064] In general, the metabolic activity vector is mapped into a metabolic pathology feature vector through a fully connected layer, and the metabolic dynamic information is efficiently integrated in the form of a structured vector, avoiding the time-consuming problem of traditional frame-by-frame analysis of time series data. Both features are quickly generated in a compact format, laying an efficient foundation for subsequent fusion operations.

[0065] In general, the fusion process of primary pathological features and metabolic pathology feature vectors is achieved through direct splicing or feature mapping of structured data, without the need for complex cross-modal alignment calculations, significantly improving the efficiency of pathology label generation.

[0066] In general, the generated pathology labels can be directly used as the root nodes of the scalable index tree. Their precise pathology semantic information can accelerate the construction of the index hierarchy, making the index tree more in line with clinical diagnosis logic, and thus achieving rapid location and retrieval of data during archiving.

[0067] In general, standardized pathology labels facilitate encrypted and fragmented storage after being bound to standard three-dimensional matrices, reducing format conversion overhead during data transmission and storage, optimizing the entire chain from feature extraction to index construction to data storage, and effectively improving the archiving efficiency of multimodal medical images.

[0068] S5. Constructing an extensible index tree for the patient based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file; In an embodiment of the present invention, constructing the patient's extensible index tree based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file includes: Generate a first level with the pathology label as the root node; Encode the device type as an orthogonal feature vector at the second level; Generate the third-level leaf nodes based on the frequency domain features of the imaging modality; An extensible index tree of the patient is constructed using the root node, the orthogonal feature vectors, and the leaf nodes.

[0069] After constructing the patient's extensible index tree based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file, the method includes: Pruning and merging the leaf nodes whose similarity is higher than the redundancy threshold; The node weights of the scalable index tree are dynamically updated based on data timeliness.

[0070] Specifically, a starting root node is first determined, a specific connection relationship is established between the orthogonal eigenvector and the root node, and the leaf nodes are connected to the corresponding positions in sequence according to certain rules. Through this hierarchical connection method, the root node, orthogonal eigenvector and leaf nodes are combined in an orderly manner to construct an extensible index tree for the patient.

[0071] Furthermore, after constructing an extensible index tree for the patient based on the pathological label, the device type of the medical imaging device, and the imaging modality in the original data file, all leaf nodes in the tree are compared one by one.

[0072] Furthermore, the similarity between every two leaf nodes is calculated, and the calculated similarity is compared with a pre-set redundancy threshold. For leaf nodes whose similarity is higher than the redundancy threshold, these similar leaf nodes are merged into one node, and redundant branches are deleted to complete the pruning and merging operation.

[0073] Furthermore, the node weights of the extensible index tree are adjusted according to the time when the data is generated and updated.

[0074] Furthermore, nodes corresponding to newly generated data are given higher weights; as time goes by, the weights of nodes corresponding to old data gradually decrease.

[0075] Furthermore, according to the changes in data timeliness, the weight of each node is regularly re-evaluated and modified, thereby realizing dynamic update of the weight of the scalable index tree nodes based on data timeliness.

[0076] In general, with the pathology label as the root node (directly related to clinical diagnosis needs at the semantic level), the device type is encoded as an orthogonal feature vector (standardizing the data attributes of different devices), and the frequency domain features of the imaging modality are used as leaf nodes (capturing the modal characteristics of the data), a hierarchical structure that conforms to the logic of medical data retrieval is formed, so that archived data can be quickly located through pathology semantics, device type or modality features, greatly shortening the retrieval time.

[0077] In general, duplicate data indexes are eliminated by pruning and merging leaf nodes with similarity above a threshold; node weights are dynamically updated based on data timeliness to increase the index priority of frequently accessed or latest data, avoiding invalid indexes from occupying computing resources and ensuring that the index tree always maintains efficient retrieval performance.

[0078] In general, after the index tree is bound to the standard three-dimensional matrix, encrypted fragmented storage can be performed directly based on the index structure (such as dividing by organ area), so that the storage location can be quickly located according to the index hierarchy when archiving data. At the same time, the archive integrity can be verified through hash value comparison, reducing the time spent on data verification and achieving efficiency improvement in the entire process from index construction to storage and archiving.

[0079] S6. Bind the standard three-dimensional matrix to the extensible index tree, and encrypt and store the fragments in the medical cloud node.

[0080] In an embodiment of the present invention, the encrypted fragments are stored in the medical cloud node, including: encrypting the bound extensible index tree using the patient's biometric key; Dividing the bound extensible index tree into blocks according to the organ regions of the standard three-dimensional matrix to obtain data blocks of the original data file; Distribute and store data blocks to different medical cloud nodes.

[0081] The step of binding the standard three-dimensional matrix to the scalable index tree and encrypting and storing the fragments in the medical cloud node includes: When the original data file is archived, restoring the data block from the medical cloud node; Reconstructing the data block encryption matrix to obtain a hash value of the archived data block; Comparing the deviation between the hash value and the original hash value of the data block to obtain the deviation value of the archived data block; When the deviation value is lower than the data damage threshold, a successful archiving signal is triggered.

[0082] Specifically, the patient's biometric key is obtained. The key is generated based on the patient's unique biometric characteristics, such as fingerprints, irises and other information that has been converted through specific processing.

[0083] Furthermore, the biometric key is used to encrypt the bound extensible index tree using a specific encryption method, and the data in the index tree is scrambled and reorganized so that people who do not hold the correct biometric key cannot read the content, thereby ensuring data security.

[0084] Furthermore, the bound extensible index tree is segmented with reference to the organ regions divided in the standard three-dimensional matrix.

[0085] Furthermore, according to the boundary range of the organ region, the data parts related to each organ region in the index tree are extracted separately to form independent data units. These data units are data blocks of the original data file, and each data block corresponds to index information related to one or more organ regions.

[0086] Furthermore, the data blocks of the original data files are stored separately on different medical cloud nodes. Each data block is sent to a pre-selected medical cloud storage server via network transmission. Each cloud node is responsible for storing a portion of the data blocks, achieving distributed data storage, ensuring data storage reliability and accessibility, while reducing storage pressure on individual nodes.

[0087] Specifically, all leaf nodes in the extensible index tree are checked one by one, and the similarity between every two leaf nodes is calculated according to a pre-set unified comparison rule.

[0088] Furthermore, the calculated leaf node similarity result is compared with a pre-determined redundancy threshold.

[0089] Furthermore, once it is found that the similarity between two or more leaf nodes is higher than the redundancy threshold, it is determined that the information contained in these leaf nodes is redundant.

[0090] Furthermore, these similar leaf nodes are merged into one node, and redundant branch connections between them are removed, thus completing the pruning and merging operation of the scalable index tree leaf nodes.

[0091] Furthermore, the basis for judging the timeliness of the data is determined, for example, based on the time when the data was generated, the closer the data is to the current time, the higher the timeliness of the data.

[0092] Furthermore, based on this judgment, all nodes in the scalable index tree are regularly checked. Nodes corresponding to newly generated data are given a higher weight to highlight their importance; over time, the weights of nodes corresponding to older data are gradually reduced.

[0093] Furthermore, according to the changes in data timeliness, the weight of each node is continuously re-evaluated and adjusted, so as to realize the dynamic update of the node weight of the scalable index tree based on data timeliness.

[0094] In general, the standard three-dimensional matrix is ​​bound to the extensible index tree to optimize archiving efficiency through the direct association between structured data and indexes: In general, the extensible index tree takes the pathology label as the root node, the device type and imaging modality as the hierarchical features, and after being bound to the standardized three-dimensional matrix, forms a direct mapping relationship of "pathology semantics-device attributes-imaging modality-data entity".

[0095] In general, when archiving, the data storage location can be quickly located through the index tree, avoiding traversing the entire data and significantly shortening data retrieval and call time.

[0096] In general, the standard three-dimensional matrix unifies the spatial coordinate system and data format of multimodal images, and does not require additional format conversion when binding with the index tree, reducing preprocessing time.

[0097] In general, the hierarchical structure of the index tree matches the organ area division of the three-dimensional matrix, making the data storage logic more in line with clinical application scenarios and improving the smoothness of the archiving process.

[0098] In general, the bound matrix is ​​divided into blocks by organ region and distributed across different cloud nodes, enabling parallel processing of data storage. Compared to traditional centralized storage, this approach can leverage the concurrent read and write capabilities of multiple nodes, significantly reducing the time cost of large-scale data archiving. It is particularly suitable for scenarios with massive multimodal imaging data.

[0099] In general, using a patient's biometric key to encrypt the index tree ensures data security while mitigating the impact of complex encryption algorithms on storage efficiency. The strategy of partitioning data by organ region, based on anatomical characteristics, aligns data block size with access frequency, further optimizing storage and retrieval efficiency.

[0100] In general, archiving quickly verifies data integrity by restoring data blocks, reconstructing the encryption matrix, and comparing hash values. This mechanism eliminates the need for byte-by-byte verification; instead, it determines data corruption simply by comparing hash value deviations. When the deviation falls below a threshold, a successful archiving signal is triggered, significantly reducing data verification time and preventing archiving process bottlenecks caused by time-consuming verification.

[0101] In general, through the integrated process of "binding-encryption-sharding-verification", the efficiency of the entire chain of multimodal image archiving can be improved.

[0102] In general, the binding of the index tree and the three-dimensional matrix reduces data retrieval latency and meets the needs of rapid clinical retrieval.

[0103] In general, distributed sharded storage reduces the load on a single node and, combined with the lightweight design of biometric encryption, reduces network transmission and storage resource consumption.

[0104] In general, the hash verification mechanism ensures the accuracy of archived data, avoids repeated processing due to data errors, and forms an efficient closed loop of "storage-verification-archiving".

[0105] like Figure 2 FIG. 1 is a functional module diagram of a multimodal medical image intelligent archiving system provided by an embodiment of the present invention.

[0106] The multimodal medical image intelligent archiving system 100 described in the present invention can be installed in an electronic device. Depending on the functionality implemented, the multimodal medical image intelligent archiving system 100 may include an image conversion module 101, an anatomical structure atlas generation module 102, a metabolic activity vector generation module 103, a pathology label generation module 104, an extensible index tree construction module 105, and a data archiving module 106. A module, also referred to as a unit, is a series of computer program segments that can be executed by an electronic device processor and perform a fixed function, and is stored in the electronic device's memory.

[0107] In this embodiment, the functions of each module / unit are as follows: The image conversion module 101 is used to dynamically convert the original data file of the medical imaging device into a standard three-dimensional matrix; The anatomical structure atlas generating module 102 is configured to extract a multi-channel gradient feature map from the standard three-dimensional matrix, perform organ segmentation on the multi-channel gradient feature map, and perform contour abnormality correction on the multi-channel gradient feature map after organ segmentation based on anatomical topological constraints to obtain an anatomical structure atlas of the patient; The metabolic activity vector generating module 103 is configured to fit the time information and shadow intensity of the dynamic image in the standard three-dimensional matrix into the metabolic curve of the patient, and encode the metabolic curve into the metabolic activity vector of the patient; The pathology label generation module 104 is configured to extract pathology abnormality features from the anatomical structure atlas, fuse the pathology abnormality features with the metabolic activity vector, and obtain the pathology label of the patient; The extensible index tree construction module 105 is configured to construct an extensible index tree for the patient based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file; The data archiving module 106 is used to bind the standard three-dimensional matrix to the scalable index tree and store the encrypted fragments in the medical cloud node. In the several embodiments provided by the present invention, it should be understood that the disclosed method and system can be implemented in other ways. For example, the system embodiment described above is merely illustrative. For example, the module division is only a logical functional division, and other division methods may be used in actual implementation.

[0108] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.

[0109] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0110] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0111] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multimodal medical image intelligent archiving method, characterized in that: The method comprises: S1. Dynamically convert the raw data files of medical imaging equipment into a standard three-dimensional matrix; S2. extracting a multi-channel gradient feature map from the standard three-dimensional matrix, performing organ segmentation on the multi-channel gradient feature map, and correcting contour anomalies on the multi-channel gradient feature map after organ segmentation based on anatomical topological constraints to obtain an anatomical structure atlas of the patient; S3. fitting the time information and shadow intensity of the dynamic image in the standard three-dimensional matrix into a metabolic curve of the patient, and encoding the metabolic curve into a metabolic activity vector of the patient; S4. extracting pathological abnormality features from the anatomical structure atlas, fusing the pathological abnormality features with the metabolic activity vector, and obtaining a pathological label for the patient; S5. Constructing an extensible index tree for the patient based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file; S6. Bind the standard three-dimensional matrix to the extensible index tree, and encrypt and store the fragments in the medical cloud node.

2. The multimodal medical image intelligent archiving method according to claim 1, wherein: The method of dynamically converting the raw data file of the medical imaging device into a standard three-dimensional matrix includes: Parsing the header information of the original data file to obtain the spatial resolution parameter of the original data file; performing isotropic resampling on the image data in the original data file according to the spatial resolution parameter; The isotropically resampled image data is mapped to a standard coordinate system to generate a standard three-dimensional matrix of the original data file.

3. The multimodal medical image intelligent archiving method according to claim 1, wherein: The step of performing organ segmentation on the multi-channel gradient feature map and correcting contour abnormalities on the multi-channel gradient feature map after organ segmentation based on anatomical topology constraints to obtain an anatomical structure atlas of the patient includes: Fusing shallow high-resolution features and deep semantic features of the multi-channel gradient feature map to obtain a patient's organ probability heat map; performing contour abnormality correction on the binary organ probability heat map based on anatomical topology constraints to obtain a binary organ mask atlas of the patient; The outer surface triangular mesh of the binary organ mask atlas is mapped back to the spatial coordinate system of the standard three-dimensional matrix to obtain the anatomical structure atlas of the patient.

4. The multimodal medical image intelligent archiving method according to claim 1, wherein: The step of fitting the time information and shadow intensity of the dynamic image in the standard three-dimensional matrix into a metabolic curve of the patient, and encoding the metabolic curve into a metabolic activity vector of the patient, comprises: Extract voxel intensity values ​​at consecutive time points in dynamic images; identifying the shadow intensity of the target organ region in the dynamic image by using the voxel intensity value; The time information and the shadow intensity are fitted into the metabolic curve of the patient, wherein the fitting calculation formula is as follows: , Where, for Plasma input at any time, for The contrast agent concentration in the target organ area is linearly converted from the shadow intensity at each moment. is the contrast agent transmembrane transport rate constant, is the tissue clearance rate constant, is the time information, is the differential symbol, is the metabolic profile of the patient; After the curvature features of the metabolic curve are spliced ​​into a high-dimensional vector, the dimension is reduced to the metabolic activity vector of the patient.

5. The multimodal medical image intelligent archiving method according to claim 1, wherein: The extracting of the abnormal pathological features of the anatomical structure atlas, fusing the abnormal pathological features with the metabolic activity vector, and obtaining the pathological label of the patient includes: Locating the abnormal area in the anatomical structure atlas, and using the morphological indexes and texture indexes of the abnormal area as primary pathological features; Mapping the metabolic activity vector into a metabolic pathology feature vector through a fully connected layer; The primary pathological feature is fused with the metabolic pathological feature vector to obtain the pathological label of the patient.

6. The multimodal medical image intelligent archiving method according to claim 1, wherein: The constructing of the patient's extensible index tree based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file includes: Generate a first level with the pathology label as the root node; Encode the device type as an orthogonal feature vector at the second level; Generate the third-level leaf nodes based on the frequency domain features of the imaging modality; An extensible index tree of the patient is constructed using the root node, the orthogonal feature vectors, and the leaf nodes.

7. The multimodal medical image intelligent archiving method according to claim 6, wherein: After constructing the patient's extensible index tree based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file, the method includes: Pruning and merging the leaf nodes whose similarity is higher than the redundancy threshold; The node weights of the scalable index tree are dynamically updated based on data timeliness.

8. The multimodal medical image intelligent archiving method according to claim 1, wherein: The encrypted shards are stored in the medical cloud node, including: encrypting the bound extensible index tree using the patient's biometric key; Dividing the bound extensible index tree into blocks according to the organ regions of the standard three-dimensional matrix to obtain data blocks of the original data file; Distribute and store data blocks to different medical cloud nodes.

9. The multimodal medical image intelligent archiving method according to claim 8, wherein: The step of binding the standard three-dimensional matrix to the scalable index tree and encrypting and storing the fragments in the medical cloud node includes: When the original data file is archived, restoring the data block from the medical cloud node; Reconstructing the data block encryption matrix to obtain a hash value of the archived data block; Comparing the deviation between the hash value and the original hash value of the data block to obtain the deviation value of the archived data block; When the deviation value is lower than the data damage threshold, a successful archiving signal is triggered.

10. A multimodal medical image intelligent archiving system, characterized in that: The system comprises: Image conversion module, used to dynamically convert raw data files of medical imaging equipment into standard three-dimensional matrices; an anatomical structure atlas generation module, configured to extract a multi-channel gradient feature map from the standard three-dimensional matrix, perform organ segmentation on the multi-channel gradient feature map, and perform contour abnormality correction on the multi-channel gradient feature map after organ segmentation based on anatomical topological constraints to obtain an anatomical structure atlas of the patient; a metabolic activity vector generation module, configured to fit the time information and shadow intensity of the dynamic image in the standard three-dimensional matrix into a metabolic curve of the patient, and encode the metabolic curve into a metabolic activity vector of the patient; a pathology label generation module, configured to extract pathology abnormality features from the anatomical structure atlas, fuse the pathology abnormality features with the metabolic activity vector, and obtain a pathology label for the patient; an extensible index tree construction module, configured to construct an extensible index tree for the patient based on the pathology label, the device type of the medical imaging device, and the imaging modality in the original data file; The data archiving module is used to bind the standard three-dimensional matrix with the extensible index tree and store the encrypted fragments in the medical cloud node.

Images