Nursing data management optimization method and system for brain tumor patient

Abstract

The invention relates to the technical field of nursing data management, and particularly discloses a nursing data management optimization method and system for a brain tumor patient, which is used for performing structured conversion and FHIR format standardization processing based on a multi-dimensional rule on vital sign data and electronic medical record data of the brain tumor patient. Therefore, the consistency of data formats is ensured. For medical image data of a patient, a large model technology is particularly introduced to carry out automatic structured description on a medical image, through image feature extraction and semantic segmentation based on deep learning, morphological features and a spatial distribution mode of a focus are identified, a structured image description text is generated, and FHIR format conversion is further completed. Finally, unified access, storage, query and analysis of vital sign data, electronic medical record records and medical image reports are achieved by deploying an integrated server meeting the FHIR standard. According to the method, the integration level of nursing data can be remarkably improved, information islands are broken, and data interoperability is improved.

Description

Technical Field

[0001] The present application relates to the technical field of nursing data management, and more specifically, to a nursing data management optimization method and system for brain tumor patients. Background Art

[0002] Brain tumors are complex and life-threatening diseases, and their treatment often involves multidisciplinary collaboration and long-term, meticulous nursing management. For patients with brain tumors, nurses need to closely monitor changes in their vital signs, assess disease progression, implement medical orders, observe treatment responses and potential complications, and provide effective health education and psychological support. This series of nursing activities is highly dependent on comprehensive, accurate, and timely access to all types of patient information, including real-time updated vital sign data, electronic medical records that record the diagnosis and treatment process, and medical images that reveal the condition of the lesion. Therefore, building an efficient and integrated nursing data management solution is of vital importance to improving the quality of care for brain tumor patients, ensuring patient safety, and optimizing the use of medical resources.

[0003] However, in current medical practice, managing nursing data for brain tumor patients often faces numerous challenges. Existing medical information systems, such as hospital information systems (HIS), electronic health records (EHR), and picture archiving and communication systems (PACS), typically operate independently, resulting in patient data being stored in separate "information silos." Vital sign data, electronic medical records, and medical imaging reports often have varying formats and standards, lacking effective interoperability. Nursing staff may need to log into multiple systems to access information, consuming significant time and effort, making it difficult to quickly form a holistic understanding of the patient's condition, thereby impacting the efficiency and accuracy of nursing decisions. Medical imaging data, in particular, is inherently unstructured visual information. Conventional PACS systems can only store and access medical images but are unable to convert lesion information in the images into structured data that can be used for clinical decision-making, limiting the efficiency of developing personalized care plans.

[0004] Therefore, a nursing data management optimization method and system for brain tumor patients is expected. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a nursing data management optimization method and system for brain tumor patients, which performs structured conversion and FHIR format standardization processing based on multi-dimensional rules such as field name mapping and coding system conversion on the vital signs data and electronic medical record data of brain tumor patients to ensure the consistency of data format. For the patient's medical imaging data, large model technology is specially introduced to perform automatic structured description of medical images. Through deep learning-based image feature extraction and semantic segmentation, the morphological characteristics and spatial distribution patterns of the lesions are identified, structured image description text is generated, and FHIR format conversion is further completed. Finally, by deploying an integrated server that complies with the FHIR standard, unified access, storage, query and analysis of vital signs data, electronic medical records and medical imaging reports are achieved. This method can significantly improve the integration of nursing data, break information silos, enhance data interoperability, and provide strong data support for the nursing management of brain tumor patients.

[0006] Accordingly, according to one aspect of the present application, a method for optimizing nursing data management for brain tumor patients is provided, comprising:

[0007] Acquiring vital sign data of a target brain tumor patient subject, and performing FHIR conversion on the vital sign data of the target brain tumor patient subject to obtain vital sign FHIR format data;

[0008] obtaining electronic medical record data of the target brain tumor patient;

[0009] Extract key data from the electronic medical record data and perform FHIR conversion using an FHIR adapter to obtain electronic medical record FHIR format data;

[0010] Acquiring medical imaging data of the target brain tumor patient;

[0011] Performing a medical image structured description based on a large model on the medical image data of the target brain tumor patient to obtain a medical image structured description, and performing FHIR conversion on the medical image structured description to obtain medical image structured description FHIR format data;

[0012] Deploy a server that complies with the FHIR standard, wherein the server that complies with the FHIR standard is used to store and manage the vital sign FHIR format data, the electronic medical record FHIR format data, and the medical image structured description FHIR format data;

[0013] The FHIR-compliant server provides a standard FHIR API for other application systems to access and query data.

[0014] According to another aspect of the present application, a nursing data management optimization system for brain tumor patients is provided, comprising:

[0015] a vital sign data FHIR conversion module, configured to obtain vital sign data of a target brain tumor patient object, and perform FHIR conversion on the vital sign data of the target brain tumor patient object to obtain vital sign FHIR format data;

[0016] an electronic medical record data acquisition module, configured to acquire electronic medical record data of the target brain tumor patient;

[0017] An electronic medical record FHIR conversion module is used to extract key data and perform FHIR conversion on the electronic medical record data through an FHIR adapter to obtain electronic medical record FHIR format data;

[0018] a medical imaging data acquisition module, configured to acquire medical imaging data of the target brain tumor patient;

[0019] a medical image structured description module, configured to perform a medical image structured description based on a large model on the medical image data of the target brain tumor patient to obtain a medical image structured description, and perform FHIR conversion on the medical image structured description to obtain medical image structured description FHIR format data;

[0020] A server deployment module is used to deploy a server that complies with the FHIR standard. The server that complies with the FHIR standard is used to store and manage the vital signs FHIR format data, the electronic medical record FHIR format data, and the medical image structured description FHIR format data. The server that complies with the FHIR standard provides a standard FHIR API for other application systems to access and query data.

[0021] Compared with the existing technology, the nursing data management optimization method and system for brain tumor patients provided by this application performs structured conversion and FHIR format standardization processing based on multi-dimensional rules such as field name mapping and coding system conversion on the vital signs data and electronic medical record data of brain tumor patients to ensure the consistency of data format. For the medical imaging data of patients, large model technology is specially introduced to perform automated structured description of medical images. Through deep learning-based image feature extraction and semantic segmentation, the morphological characteristics and spatial distribution patterns of lesions are identified, structured image description text is generated, and FHIR format conversion is further completed. Finally, by deploying an integrated server that complies with the FHIR standard, unified access, storage, query and analysis of vital signs data, electronic medical records and medical imaging reports are achieved. This method can significantly improve the integration of nursing data, break information silos, enhance data interoperability, and provide strong data support for the nursing management of brain tumor patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0023] Figure 1 The flowchart is a method for optimizing nursing data management for brain tumor patients according to an embodiment of the present application.

[0024] Figure 2 This is a flowchart of step S3 in the nursing data management optimization method for brain tumor patients according to an embodiment of the present application.

[0025] Figure 3 This is a flowchart of step S5 in the nursing data management optimization method for brain tumor patients according to an embodiment of the present application.

[0026] Figure 4 Schematic diagram of data flow in step S5 of the nursing data management optimization method for brain tumor patients according to an embodiment of the present application.

[0027] Figure 5 Flowchart of step S52 in the nursing data management optimization method for brain tumor patients according to an embodiment of the present application.

[0028] Figure 6 4 is a block diagram of a nursing data management optimization system for brain tumor patients according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0030] Figure 1 FIG. 1 is a flow chart of a nursing data management optimization method for brain tumor patients according to an embodiment of the present application. Figure 1 As shown, according to an embodiment of the present application, the nursing data management optimization method for brain tumor patients includes the following steps: S1, obtaining vital sign data of a target brain tumor patient object, and performing FHIR conversion on the vital sign data of the target brain tumor patient object to obtain vital sign FHIR format data; S2, obtaining electronic medical record data of the target brain tumor patient object; S3, extracting key data and performing FHIR conversion on the electronic medical record data through an FHIR adapter to obtain electronic medical record FHIR format data; S4, obtaining medical imaging data of the target brain tumor patient object; S5, obtaining medical imaging data of the target brain tumor patient object; S6, obtaining medical imaging data of the target brain tumor patient object; S7, obtaining medical imaging data of the target brain tumor patient object; S8, obtaining medical imaging data of the target brain tumor patient object; S9, obtaining medical imaging data of the target brain tumor patient object; S10, obtaining medical imaging data of the target brain tumor patient object; S11, obtaining medical imaging data of the target brain tumor patient object; S12, obtaining medical imaging data of the target brain tumor patient object; S13, obtaining medical imaging data of the target brain tumor patient object; S14, obtaining medical imaging data of the target brain tumor patient object; S15, obtaining medical imaging data of the target brain tumor patient object; S16, obtaining medical imaging data of the target brain tumor patient object; 5. Perform a large-scale model-based medical image structured description on the medical image data of the target brain tumor patient to obtain a medical image structured description, and perform FHIR conversion on the medical image structured description to obtain medical image structured description FHIR format data; S6. Deploy a server that complies with the FHIR standard, and the server that complies with the FHIR standard is used to store and manage the vital sign FHIR format data, the electronic medical record FHIR format data, and the medical image structured description FHIR format data; S7. The server that complies with the FHIR standard provides a standard FHIR API for other application systems to access and query data.

[0031] In the above-mentioned nursing data management optimization method for brain tumor patients, the step S1 obtains the vital signs data of the target brain tumor patient object, and performs FHIR conversion on the vital signs data of the target brain tumor patient object to obtain vital signs FHIR format data. It should be understood that the vital signs of brain tumor patients (such as blood pressure, heart rate, respiration, body temperature, blood oxygen saturation, etc.) are key indicators for evaluating changes in the condition and nursing effects, but these data often come from different monitoring devices or recording systems, with non-uniform formats, and are difficult to integrate and utilize. Therefore, in order to achieve the standardization of vital signs data and facilitate subsequent unified storage, query and correlation analysis with other clinical data, this application obtains the original vital signs data and performs FHIR conversion on it to obtain vital signs data in a unified format. Specifically, the FHIR (Fast Healthcare Interoperability Resources) standard provides a set of well-defined medical data resource models and exchange specifications based on modern Web technology, which includes, for example, Observation resources, which can standardize the representation of vital signs measurement results. In specific implementation, vital sign data (such as blood pressure 120 / 80 mmHg, heart rate 75 beats / minute, etc.) is collected from monitoring devices, HIS systems, or nursing records through the system interface, and key information (measurement item, value, unit, timestamp, patient identification) is extracted. Then, according to the specifications of the FHIR Observation resource, this information is mapped and filled into the corresponding fields (such as code for measurement item, valueQuantity for value and unit, effectiveDateTime for measurement time, and subject for patient resource association), and finally JSON or XML format data that complies with the FHIR standard is generated. In this way, the original vital sign data in various formats is converted into standard and unified vital sign FHIR format data, laying the foundation for cross-system interoperability and integrated management.

[0032] In the above-mentioned nursing data management optimization method for brain tumor patients, the step S2 is to obtain the electronic medical record data of the target brain tumor patient object. It should be understood that the electronic medical record (EHR) contains rich and critical clinical information such as the patient's diagnosis information, treatment plan, medication records, nursing records, laboratory test results, etc., and is the core data source for a comprehensive understanding of the patient's health status and diagnosis and treatment process. Therefore, in order to be able to extract and utilize these important clinical context information to support more comprehensive nursing data management and analysis, this application connects and obtains the electronic medical record data of the target patient to achieve the integration of electronic medical record data. In the specific implementation, the hospital's electronic medical record system is accessed through a pre-configured interface (such as database connection, Web service call, HL7 message interface, etc.), and the patient's unique identifier (such as hospitalization number, ID number, etc.) is used to query and retrieve the relevant medical record documents or data.

[0033] In the above-mentioned nursing data management optimization method for brain tumor patients, the step S3 extracts key data and performs FHIR conversion on the electronic medical record data through the FHIR adapter to obtain electronic medical record FHIR format data. It should be understood that the original electronic medical record data is usually complex in structure and content, and may contain a large amount of unstructured text, which is inefficient to use directly, and its internal data model and coding system (such as drug coding, diagnosis coding) are often different from the standardized FHIR resource model. Therefore, in order to extract the most valuable key information for nursing management from the complex electronic medical records and convert it into the standardized FHIR format to achieve seamless integration and interoperability with other data, this application further uses the FHIR adapter to extract key data and perform FHIR conversion on the electronic medical record data. Here, the FHIR adapter serves as a middleware component, responsible for parsing the format and content of the original electronic medical record data, identifying and extracting key clinical information based on preset mapping rules and conversion logic, and organizing and encapsulating it according to the FHIR standard resource model, ultimately generating electronic medical record data that conforms to the FHIR standard (such as Condition resources representing diagnostic information, MedicationRequest resources representing medication orders, and Procedure resources representing surgical or operation records, etc.).

[0034] Figure 2 FIG. 1 is a flow chart of step S3 in the nursing data management optimization method for brain tumor patients according to an embodiment of the present application. Figure 2As shown, step S3 includes: S31, extracting key data elements from the electronic medical record data to obtain a set of key data elements; S32, based on data mapping rules, mapping each key data element in the set of key data elements to key data element FHIR format data to obtain a set of key data element FHIR format data; S33, performing data verification and quality control on the set of key data element FHIR format data to obtain the electronic medical record FHIR format data.

[0035] Specifically, step S31 extracts key data elements from the electronic medical record data to obtain a set of key data elements. Specifically, because raw electronic medical record (EHR) data typically contains a vast amount of information about a patient since admission, its structure is complex and diverse, including both structured tabular data (such as laboratory results) and a large amount of unstructured text (such as medical records and nursing notes). Not all information is equally important to current brain tumor nursing management goals, and directly processing the full amount of raw data is inefficient and lacks focus. Therefore, in order to focus on the most relevant information for nursing decision-making, disease monitoring, and risk assessment of brain tumor patients and improve the efficiency and accuracy of subsequent processing, this application accurately identifies and extracts pre-defined key data elements from the complete electronic medical record data, thereby filtering out high-value information subsets from the complex raw data, such as basic patient information, brain tumor-related diagnostic codes and descriptions, key neurological assessment results (such as changes in GCS scores), abnormal vital signs, special medications (such as mannitol, hormones, anti-epileptic drugs), allergy history, important laboratory indicators (such as electrolytes, coagulation function), and care-related plans and measures. Specifically, the FHIR adapter is configured with corresponding extraction rules or calls specific modules. For structured data, it's possible to directly retrieve the value of a specified field from the EHR database or interface by configuring an SQL query or API call. For unstructured or semi-structured text (such as nursing records and physician notes), natural language processing (NLP) techniques, such as named entity recognition (NER), can be applied to locate and extract key concepts mentioned (such as "intensified headache," "impaired consciousness," "drainage volume," etc.) and their associated attributes (time, degree, etc.). In this way, a preliminarily screened set of key data elements closely related to brain tumor care is prepared for subsequent standardized conversion, removing a large amount of irrelevant information and making subsequent mapping more focused and efficient.

[0036] Specifically, in step S32, based on the data mapping rules, each key data element in the set of key data elements is mapped to key data element FHIR format data to obtain a set of key data element FHIR format data. It should be understood that the data structure, naming method, coding system (such as local diagnosis code, drug code) and data type of each key data element in the set of key data elements still follow the internal specifications of the source EHR system, and there are differences with other systems or standards (especially FHIR standards), and interoperability cannot be achieved directly. Therefore, in order to convert the key information in these source system formats into a globally universal and standardized FHIR format, and ensure that the data can be accurately understood and seamlessly integrated between different systems, this application converts each key data element in the set by applying predefined detailed data mapping rules to convert the source data representation into the target FHIR resource representation. In a specific example of the present application, the data mapping rules include mapping of field names (source field name -> FHIR resource attribute path), conversion of data types (such as string date -> FHIR dateTime type), conversion of coding systems (such as local drug code -> RxNorm or national drug code; local diagnosis code -> ICD-10 or SNOMED CT) and conversion of data values ​​(such as unit conversion). Specifically, the FHIR adapter traverses each item in the key data element set. For example, the extracted diagnostic information "glioblastoma" (local code D001), the mapping rule will indicate that it should be converted into a FHIR Condition resource, where the Condition.code.coding array should contain a system of http: / / snomed.info / sct (representing the SNOMED CT coding system), code of 254837009 (SNOMED CT code for glioblastoma), and display of "Glioblastoma". Similarly, the extracted medication "Dexamethasone 5mg QD" is mapped to an FHIRMedicationRequest resource based on the rules, containing standardized information such as the medication code, dosage, frequency, and route of administration. By executing this rule on all key elements in the collection, key data elements, originally formatted in various ways and using local terminology, are converted into structured data that conforms to the FHIR standard. This forms a collection of key data elements in FHIR format, laying the foundation for the final integration into a complete FHIR representation of the electronic medical record.

[0037] Specifically, in step S33, data verification and quality control are performed on the set of key data element FHIR format data to obtain the electronic medical record FHIR format data. It should be understood that although the data mapping process is carried out in accordance with the rules, quality problems of the source data (such as missing, incorrect, inconsistent) or imperfect mapping rules may cause the generated FHIR format data fragments to still have problems, such as the lack of mandatory fields required by the FHIR specification, data type mismatch, invalid coding, or logical contradictions between data. Once these low-quality data are stored in the server, they will affect the accuracy and reliability of subsequent applications. Therefore, in order to ensure that the electronic medical record data finally stored in the FHIR server is complete, accurate, valid and fully compliant with the FHIR standard, this application further performs data verification on the set of key data element FHIR format data obtained after mapping to ensure data quality and system robustness. Specifically, the FHIR adapter or the validation service it calls will utilize officially provided FHIR validation tools or libraries to check each FHIR-formatted data fragment in the collection against the FHIR base specification and any specific implementation guides (profiles, such as the US Core Data Interoperability Specification). Verification includes structural conformance (whether it adheres to the resource definition), cardinality constraints (whether field occurrences meet requirements), data type checks, term binding (whether the coded value belongs to the specified ValueSet), and referential integrity (whether referenced resources exist). Furthermore, quality control can include business logic checks, such as validating value ranges and logical date order. For data that fails validation, different policies can be implemented based on the configuration, including logging an error, rejecting the data, attempting automatic repair (if the rules are clear), or flagging it for manual review. In this way, qualified FHIR data that has passed validation and quality control screening is assembled into a complete FHIR resource, resulting in high-quality FHIR-formatted electronic medical record data that is then submitted to the FHIR server for storage and management.

[0038] In the above-mentioned nursing data management optimization method for brain tumor patients, the step S4 obtains the medical imaging data of the target brain tumor patient object. It should be understood that since medical imaging (such as CT, MRI) is one of the "gold standards" for diagnosing brain tumors, evaluating tumor size, location, morphology and treatment response, it contains vital visual information and plays an irreplaceable role in formulating accurate nursing plans (such as observing intracranial pressure changes, position management, determining skin care areas after radiotherapy, etc.). Therefore, the present application obtains the medical imaging data of the target patient in order to achieve in-depth analysis and structured processing of the patient's medical imaging data, and extracts imaging features (such as tumor size, morphology, location, etc.) that support clinical decision-making. Specifically, the relevant brain imaging examinations of the target patient (such as MRI or CT images of a specified sequence) can be queried and retrieved from the hospital's image archiving and communication system (PACS) through the DICOM protocol (such as using C-FIND, C-MOVE commands) or other interfaces provided by the PACS system, usually obtained in the form of a DICOM format file set.

[0039] In the above-mentioned nursing data management optimization method for brain tumor patients, the step S5 is to perform a medical image structured description based on a large model on the medical image data of the target brain tumor patient object to obtain a medical image structured description, and to perform FHIR conversion on the medical image structured description to obtain medical image structured description FHIR format data. It should be understood that since traditional medical image data exists in the form of a pixel matrix, the clinical information contained therein requires a radiologist to read and write an unstructured text report before it can be used. It is difficult to achieve automated processing, accurate quantitative analysis, and deep integration with other clinical data, which limits the potential of image information in real-time, refined nursing decision support. Therefore, in order to directly apply this application, the key structured information about brain tumors (such as precise location, volume, morphological characteristics, etc.) about the medical image data is objectively extracted from the image by performing a medical image structured description based on a large model, thereby deepening the application of image data in nursing management, and further performing FHIR conversion on the extracted medical image structured description to integrate it into a unified data framework.

[0040] Figure 3 This is a flowchart of step S5 in the nursing data management optimization method for brain tumor patients according to an embodiment of the present application. Figure 4 FIG. 5 is a data flow diagram of step S5 in the nursing data management optimization method for brain tumor patients according to an embodiment of the present application. Figure 3 and Figure 4As shown, the step S5 includes: S51, extracting the lesion visual features of the medical imaging data of the target brain tumor patient object to obtain a set of lesion local visual feature coding maps; S52, performing visual feature space saliency on each lesion local visual feature coding map in the set of lesion local visual feature coding maps to obtain a set of lesion local visual feature saliency coding maps; S53, fusing the set of lesion local visual feature saliency coding maps to obtain a lesion local visual feature aggregate coding map; S54, performing image semantic segmentation on the lesion local visual feature aggregate coding map to obtain an image semantic segmentation result; S55, representing the image semantic segmentation result in a structured format to obtain a structured description of the medical image.

[0041] Specifically, step S51 extracts lesion visual features from the medical imaging data of the target brain tumor patient to obtain a set of lesion local visual feature coding maps. In a specific example of the present application, step S51 includes: performing image denoising on the medical imaging data of the target brain tumor patient to obtain denoised medical imaging data of the target brain tumor patient; and performing local image feature extraction on the denoised medical imaging data of the target brain tumor patient using a lesion visual feature extractor based on a deep separable convolutional neural network to obtain a set of lesion local visual feature coding maps.

[0042] It should be understood that since the original medical images (such as MRI, CT) may be affected by factors such as equipment limitations, environmental interference or slight movements of the patient during the acquisition process, noise will inevitably be introduced, which manifests as random pixel (or voxel) intensity fluctuations in the image, which may mask important diagnostic information such as the subtle boundaries and internal textures of brain tumors, and interfere with the accuracy of subsequent automated analysis. Therefore, in order to reduce the negative impact of noise on model judgment, this application applies an image denoising algorithm to the acquired original medical image data to improve image quality and provide a more reliable data basis for subsequent image processing. In an embodiment of the present application, median filtering and non-local mean filtering are used to process medical image data. Among them, median filtering is particularly effective for removing salt and pepper noise, while non-local mean filtering can better retain the detailed information of the image while removing Gaussian noise. By combining the two filtering methods, noise from different sources can be effectively suppressed to obtain clearer and more accurate brain tumor images.

[0043] Secondly, since the visual features (such as texture and shape) of the lesion area in the original medical image are often hidden in the complex brain tissue structure, it is difficult to accurately locate the key features by directly analyzing the global image. Therefore, in order to extract the local detail features of the lesion from the denoised medical image data, the present application further performs image segmentation on the medical image data of the target brain tumor patient after denoising, and performs a lesion visual feature extractor based on a deep separable convolutional neural network on each image block to capture local visual patterns related to brain tumors. Specifically, the deep separable convolutional neural network can retain the spatial feature extraction capability while reducing the amount of computation through lightweight convolution operations (separating spatial convolution and channel convolution), automatically learning the subtle texture, edge and shape features of the lesion (such as the spotted texture of microbleeding points and the infiltrative edge of glioma), and obtaining a set of local visual feature coding maps of the lesion, thereby achieving refined extraction of local features of the lesion and providing rich detail feature support for subsequent quantitative analysis.

[0044] Specifically, in step S52, each lesion local visual feature coding map in the set of lesion local visual feature coding maps is subjected to visual feature spatial saliency processing to obtain a set of lesion local visual feature saliency coding maps. That is, considering that the lesion local visual feature coding map contains a large number of irrelevant features from normal brain tissue, background or other non-lesion areas, which may cause the characteristic signal that truly reflects the characteristics of the lesion to be weakened, therefore, in order to enhance the feature expression related to the brain tumor area, the present application further performs visual feature spatial saliency processing on each lesion local visual feature coding map, and for each pixel position on the lesion local visual feature coding map, based on its own feature information, combines the contextual information of the neighborhood to enhance the feature expression, so that it more accurately expresses the essential visual characteristics of the brain tissue, thereby improving the accuracy and robustness of subsequent lesion recognition.

[0045] Figure 5 FIG. 5 is a flow chart of step S52 in the nursing data management optimization method for brain tumor patients according to an embodiment of the present application. Figure 5As shown, the step S52 includes: S521, extracting the channel feature vector of the (i, j)th pixel position from the lesion local visual feature coding map as the local lesion pixel-level visual feature vector to be enhanced; S522, based on the feature distribution of the lesion local pixel-level visual feature vector to be enhanced, performing local receptive field anchoring on the lesion local visual feature coding map to screen out a set of lesion local pixel-level visual feature vectors within the local receptive field; S523, based on the set of lesion local pixel-level visual feature vectors within the local receptive field, performing feature saliency guided by context feature association on the lesion local pixel-level visual feature vector to be enhanced to obtain a lesion local pixel-level visual feature salient coding vector, wherein the lesion local pixel-level visual feature salient coding vector is the channel feature vector of the (i, j)th pixel position in the lesion local visual feature salient coding map.

[0046] More specifically, the step S521 can be expressed as follows:

[0047] F∈R H×W×C

[0048] v tbs =v i，j ∈R C

[0049] Where R represents a real number set, H, W and C represent the height, width and number of channels of the local feature map of the lesion, F represents the local feature map of the lesion, v i，j represents the channel feature vector of the (i, j)th pixel position in the local feature map of the lesion, v tbs Represents the pixel-level visual feature vector of the local lesion to be enhanced.

[0050] Here, in order to establish an anchor point for local context analysis and feature enhancement, the present application extracts the channel feature vector of the (i, j)th pixel position from the local visual feature encoding map of the lesion as an instantiated processing unit to perform subsequent feature space saliency operations.

[0051] More specifically, step S522 includes: first, performing feature decoupling on the lesion local feature map along the channel dimension to obtain a set of lesion local pixel-level visual feature vectors, which can be expressed as:

[0052]

[0053] Among them, v 1，1 、v 1，W 、v H，1 and v H，Wdenote the local pixel-level visual feature vectors of the lesion at the (1,1), (1,W), (H,1) and (H,W) pixel positions of the lesion local feature map, respectively. FeatureDecoupling(·) denotes the feature decoupling function.

[0054] It should be understood that since the local feature map of the lesion is coupled in the channel dimension, it is not conducive to a refined analysis of the relationship between each pixel and its neighborhood. Therefore, in order to break the direct dependence between channels and reduce channel coupling, this application decouples the local feature map of the lesion along the channel dimension to convert the local feature map of the lesion into a set consisting of H×W independent pixel-level feature vectors of dimension C, thereby facilitating subsequent operations on a single pixel and its neighborhood, and mining more fine-grained feature interaction information.

[0055] Next, the pixel-level visual feature vector of the lesion to be enhanced is compressed to obtain a compressed coding vector of the pixel-level visual feature of the lesion to be enhanced, which is expressed as follows:

[0056]

[0057] Among them, ‖·‖ represents the calculation norm, v s Represents the compressed encoding vector of the pixel-level visual features of the local lesion to be enhanced.

[0058] Here, considering that the pixel-level visual feature vector of the local lesion to be enhanced may have information redundancy, which affects the efficiency of feature utilization, in order to reduce the dimension or reduce the information redundancy and extract key information, this application compresses the information of the pixel-level visual feature vector of the local lesion to be enhanced, extracts key information and removes redundancy, so as to improve the efficiency and accuracy of feature representation and provide a more compact and effective feature representation for subsequent processing.

[0059] Then, based on the characteristic distribution spatial structure characteristics of the compressed coding vector of the pixel-level visual features of the local lesion to be enhanced, the size of the characteristic receptive field of the compressed coding vector of the pixel-level visual features of the local lesion to be enhanced is determined, which can be expressed as follows:

[0060]

[0061] Wherein, log2(·) represents the logarithmic function with base 2, and r represents the half-side length of the feature receptive field of the compressed encoding vector of the pixel-level visual features to be enhanced in the local area of ​​the lesion.

[0062] Due to the different local structural complexity and information density of different image regions, using a fixed receptive field to aggregate contextual information may not be optimally adapted to all situations (for example, a large receptive field may be required for a flat area, and a small receptive field may be required for an area with complex textures), which limits the effectiveness of feature enhancement. Therefore, in order to achieve adaptive adjustment of the receptive field size, this application performs a feature distribution spatial structure analysis on the pixel-level visual feature compression coding vector of the local lesion to be enhanced, so as to dynamically determine the optimal receptive field size based on its information density, thereby ensuring that the local neighborhood contextual information can be more effectively utilized in the feature saliency process and reducing the interference of irrelevant information.

[0063] Furthermore, based on the size of the feature receptive field, a set of local pixel-level visual feature vectors of the lesion within the local receptive field is filtered out from the set of local pixel-level visual feature vectors of the lesion, which can be expressed as follows:

[0064]

[0065] Among them, v i-r，j 、v i+r，j 、v i，j+r and v i+r，j+r They respectively represent the local pixel-level visual feature vectors of the lesion at the (ir,j), (i+r,j), (i,j+r) and (i+r,j+r) positions in the local feature map of the lesion, and W represents the set of local pixel-level visual feature vectors of the lesion in the local receptive field.

[0066] That is, a matrix range is constructed with a determined receptive field size, and the pixel-level visual feature vectors therein are screened. Specifically, this application takes the location of the local pixel-level visual feature vector of the lesion to be enhanced as the center, delimits the matrix range according to the determined receptive field size, and screens out the local pixel-level visual feature vectors of the lesion within the range to construct a contextual information library containing information such as surrounding tissue texture and edges, providing a basis for subsequent saliency coding.

[0067] More specifically, step S523 includes: first, based on the feature correlation of each local pixel-level visual feature vector of the lesion in the local receptive field with respect to the local pixel-level visual feature vector of the lesion to be enhanced, performing significance aggregation on the set of local pixel-level visual feature vectors of the lesion in the local receptive field to obtain a significance aggregation coding vector of the local pixel-level visual feature of the lesion in the local receptive field, which is expressed as follows:

[0068]

[0069] Among them, v m，nrepresents the local pixel-level visual feature vector of the lesion in the local receptive field at the (m, n)th position in the set of local pixel-level visual feature vectors of the lesion in the local receptive field, (·) T represents the transpose of a vector, represents vector multiplication, softmax(·) represents the softmax normalization function, θ(v m，n ) indicates v m，n The significance weight of V s Represents the saliency aggregation encoding vector of the local pixel-level visual features of the lesion within the local receptive field.

[0070] That is, since all local pixel-level visual feature vectors of lesions within the receptive field are not equally important, the neighborhood features that are more relevant to the central pixel feature (i.e., the local pixel-level visual feature vector of the lesion to be enhanced) should have a greater weight during aggregation, so as to more effectively enhance the significance expression of the central feature. Therefore, the present application further calculates the feature correlation of the local pixel-level visual feature vector of the lesion in each local receptive field relative to the local pixel-level visual feature vector of the lesion to be enhanced, and uses it as the set aggregation weight to selectively and weightedly fuse the context information, ensuring that the neighborhood features that are highly correlated with the central pixel feature are emphasized, thereby obtaining a significant aggregation representation of the neighborhood context information, that is, the significant aggregation coding vector of the local pixel-level visual feature of the lesion in the local receptive field.

[0071] Next, the significant aggregated coding vector of the local pixel-level visual feature of the lesion in the local receptive field and the local pixel-level visual feature vector of the lesion to be enhanced are fused to obtain the significant coding vector of the local pixel-level visual feature of the lesion, which can be expressed as follows:

[0072] v' i，j =α·v i，j +β·V s

[0073] Among them, α and β represent the weighting coefficients of the local pixel-level visual feature vector to be enhanced and the saliency aggregation coding vector of the local pixel-level visual feature of the lesion in the local receptive field, respectively, v' i，j Represents the significant encoding vector of the local pixel-level visual features of the lesion.

[0074] That is, through the weighted aggregation operation, the own information of the local pixel-level visual feature vector to be enhanced of the lesion is effectively integrated with the neighborhood context information that has been screened, aggregated and importance considered, so as to enhance the expression of the central pixel feature based on the neighborhood context information, so that it can more accurately reflect the local essential characteristics of the lesion area, and obtain the significant coding vector of the local pixel-level visual feature of the lesion, providing a richer and more reliable feature basis for subsequent tasks such as lesion positioning.

[0075] In particular, in a preferred example of the present application, the local pixel-level visual feature saliency aggregate coding vector of the lesion in the local receptive field and the local pixel-level visual feature vector of the lesion to be enhanced are fused to obtain the local pixel-level visual feature saliency coding vector of the lesion, including: modulating the weighting coefficients of the local pixel-level visual feature saliency aggregate coding vector of the lesion in the local receptive field and the local pixel-level visual feature vector of the lesion to be enhanced based on the conformal fusion regularity constraint of the set of local pixel-level visual feature vectors of the lesion in the local receptive field; and performing weighted fusion on the local pixel-level visual feature saliency aggregate coding vector of the lesion in the local receptive field and the local pixel-level visual feature vector of the lesion to be enhanced based on the weighting coefficient to obtain the local pixel-level visual feature saliency coding vector of the lesion.

[0076] That is, in order to enhance the amplification effect of the feature components related to the central pixel feature significance in the set of local pixel-level visual feature vectors of the lesion within the local receptive field, it is expected that the set of pixel-level initial feature vectors within the local receptive field can have conformal representation, that is, there can be a high correspondence between its internal spatial product representation and contour gradient representation.

[0077] Therefore, first determine the internal spatial product representation vector of the set of local pixel-level visual feature vectors of the lesion in the local receptive field as:

[0078]

[0079] Where sin(·) represents the sine function, v m，n (3) An internal spatial product representation vector representing a set of local pixel-level visual feature vectors of the lesion within the local receptive field;

[0080] Next, the contour gradient vector of the set of local pixel-level visual feature vectors of the lesion in the local receptive field is determined as:

[0081]

[0082] Then, by modulating the weighting coefficients α and β, the bounded mixed tensor has an exchange regularity condition, that is, the internal spatial product representation vector v of the set of local pixel-level visual feature vectors of the lesion in the local receptive field is made m，n (3) and the contour gradient vector v m，n (2) The spatial binorm representation of the exchanged difference vectors approaches the product of coefficients α and β:

[0083] ‖v m，n (3) -v m，n(2) ‖2=ω×α×β

[0084] Where ω is the scaling factor.

[0085] In this way, by appropriately selecting the interface constraint criterion and utilizing the conformal commutativity of the inner product, the regularity specification standard is met, thereby achieving a highly conformal representation fusion of the local pixel-level visual feature vectors of each lesion within the local receptive field, thereby improving the context-aware feature significance expression of the local pixel-level visual feature vector to be enhanced.

[0086] Specifically, the step S53 is to fuse the set of the lesion local visual feature significant coding map to obtain the lesion local visual feature aggregation coding map. It should be understood that the lesion local visual feature significant coding map of a single local area only reflects the characteristics of a certain part of the lesion, and cannot form a complete description of the overall shape and spatial distribution of the lesion. Therefore, the application further integrates the respective lesion local visual feature significant coding maps into the original image space position according to the inverse process of image segmentation, so as to integrate the local features of multiple regions of the image data, construct the global feature representation of the lesion, and obtain the lesion local visual feature aggregation coding map. The lesion local visual feature aggregation coding map not only fuses the feature information from different local regions, but also retains the spatial distribution characteristics of the lesion, thereby realizing a refined description of the overall shape and spatial distribution of the lesion.

[0087] Specifically, the step S54 performs image semantic segmentation on the lesion local visual feature aggregation coding map to obtain an image semantic segmentation result. That is, in order to demarcate the boundary and range of the brain tumor in the original medical imaging data image space and realize pixel (or voxel) level distinction, the present application further performs image semantic segmentation processing on the lesion local visual feature aggregation coding map, and assigns a category label (for example, "tumor" or "background") to each pixel therein to achieve accurate identification and segmentation of the brain tumor area. In an embodiment of the present application, a U-Net network structure is used for image semantic segmentation. The U-Net network, with its unique encoder-decoder architecture, can efficiently capture contextual information in the image and retain fine spatial details at the same time, thereby accurately distinguishing brain tumor areas from normal brain tissue at the pixel level. The encoder part of the U-Net network gradually extracts high-level semantic features of the image through multiple convolution and pooling operations, while the decoder is responsible for gradually upsampling the low-resolution, high-semantic feature map extracted by the encoder back to the original image resolution, and combining the features of different levels in the encoder path (obtaining detailed information through skip connections) to predict the category probability of each pixel. It then outputs a multi-category segmentation mask of the same size as the original medical image data through thresholding as the image semantic segmentation result.

[0088] Specifically, the step S55 represents the image semantic segmentation result in a structured format to obtain the structured description of the medical image. It should be understood that although the image semantic segmentation result (tumor mask) accurately locates the tumor, it is still an image format (a collection of pixels / voxels), which is not convenient for direct data query and statistical analysis, and is difficult to directly integrate into standardized clinical data models such as FHIR. Therefore, in order to convert the spatial positioning information generated by the segmentation into quantifiable, queryable, and standardized structured data so that it can be easily understood and used by other systems, the present application further performs post-processing and information extraction on the image semantic segmentation result, extracts a series of clinically concerned indicators, and converts the geometric and spatial information contained therein into specific numerical values ​​and text descriptions. For example: calculate the number of pixels / voxels covered by the mask, and combine the pixel / voxel spacing in the image metadata to calculate the volume of the tumor (such as xx cubic centimeters); find the maximum span of the mask in three dimensions to obtain the maximum diameter of the tumor (such as xx millimeters); calculate the coordinates of the center of mass of the mask to determine the approximate position of the tumor in the anatomical space (possibly combined with standard brain atlas for alignment and then reporting the anatomical partitioning); analyze the shape characteristics of the mask (such as calculating the surface area / volume ratio, sphericity, etc. to evaluate shape regularity). Then, based on predefined templates and terminology sets, combined with NLP generation technology to automatically fill in the template, the extracted quantitative indicators and location-based descriptive information (such as "located in the left frontal lobe") are organized into a structured descriptive text to obtain a structured description of the medical image.

[0089] In the above-mentioned nursing data management optimization method for brain tumor patients, the step S6 deploys a server that complies with the FHIR standard, and the server that complies with the FHIR standard is used to store and manage the vital signs FHIR format data, the electronic medical record FHIR format data and the medical image structured description FHIR format data. It should be understood that although the various types of data (vital signs, EHR, structured images) after FHIR conversion are in a unified format, they still need a centralized and standardized place to be stored and managed in order to break the original system barriers and achieve true data convergence. Therefore, this application deploys a server that complies with the FHIR standard to store and manage all converted FHIR format data to build a unified and standard brain tumor patient care data center to facilitate subsequent data access and application. Here, the FHIR server is not just a database. It has built-in support for FHIR resource models, interaction specifications, and validation rules, and is the core infrastructure for realizing FHIR ecological interoperability. In the specific implementation, select and deploy a FHIR server software (such as HAPIFHIR, Firely Server, etc.), perform the necessary configuration, and then submit and store the vital signs FHIR format data, electronic medical record FHIR format data, and medical image structured description FHIR format data generated in the previous steps to the server through the standard FHIR RESTful API (such as POST request). The server will automatically verify, index, and manage the data, and eliminate information silos by physically integrating key data from different sources.

[0090] In the above-mentioned nursing data management optimization method for brain tumor patients, in step S7, the server that complies with the FHIR standard provides a standard FHIR API for other application systems to access and query data. That is, in order to achieve data sharing and reuse, support the development and operation of upper-layer applications, and improve overall medical care efficiency, the server that complies with the FHIR standard defined in this application has the ability to provide a standard FHIR API. It should be understood that the FHIR API is a set of standardized interfaces based on HTTP (including query, read, create, update and other operations) that allow authorized client applications to interact with the server in a unified manner to request and exchange FHIR resources. Specifically, the deployed FHIR server naturally supports the FHIR RESTful API. For example, a nursing application can query a specific patient's vital sign history, a diagnostic report containing structured image descriptions, etc. by sending a GET request to the FHIR server address. In this way, applications only need to follow the FHIR API specification to communicate with the server, without having to worry about the specific storage details of the underlying data or the differences in the source system. This greatly simplifies the access and integration process of external application systems to the comprehensive care data of brain tumor patients, promotes the flow and value realization of data, and ultimately enables smarter and more efficient patient care.

[0091] In summary, according to the embodiment of the present application, the nursing data management optimization method for brain tumor patients is explained, which performs structured conversion and FHIR format standardization processing based on multi-dimensional rules such as field name mapping and coding system conversion for the vital signs data and electronic medical record data of brain tumor patients to ensure the consistency of data format. For the patient's medical imaging data, large model technology is specially introduced to perform automated structured description of medical images. Through deep learning-based image feature extraction and semantic segmentation, the morphological characteristics and spatial distribution patterns of the lesions are identified, structured image description text is generated, and the FHIR format conversion is further completed. Finally, by deploying an integrated server that complies with the FHIR standard, unified access, storage, query and analysis of vital signs data, electronic medical records and medical imaging reports are achieved. This method can significantly improve the integration of nursing data, break information silos, enhance data interoperability, and provide strong data support for the nursing management of brain tumor patients.

[0092] Furthermore, the present application also provides a nursing data management optimization system for brain tumor patients.

[0093] Figure 6 FIG. 1 is a block diagram of a nursing data management optimization system for brain tumor patients according to an embodiment of the present application. Figure 6As shown, according to an embodiment of the present application, the nursing data management optimization system 100 for brain tumor patients includes: a vital sign data FHIR conversion module 110, which is used to obtain the vital sign data of a target brain tumor patient object and perform FHIR conversion on the vital sign data of the target brain tumor patient object to obtain vital sign FHIR format data; an electronic medical record data acquisition module 120, which is used to obtain the electronic medical record data of the target brain tumor patient object; an electronic medical record FHIR conversion module 130, which is used to extract key data and perform FHIR conversion on the electronic medical record data through an FHIR adapter to obtain electronic medical record FHIR format data; a medical image data acquisition module 140, which is used to obtain the target Medical imaging data of a brain tumor patient; a medical imaging structured description module 150, configured to perform a large-scale model-based medical imaging structured description on the medical imaging data of the target brain tumor patient to obtain a medical imaging structured description, and to perform FHIR conversion on the medical imaging structured description to obtain medical imaging structured description FHIR format data; a server deployment module 160, configured to deploy a server compliant with the FHIR standard, wherein the server compliant with the FHIR standard is configured to store and manage the vital sign FHIR format data, the electronic medical record FHIR format data, and the medical imaging structured description FHIR format data, wherein the server compliant with the FHIR standard provides a standard FHIR API for other application systems to access and query data.

[0094] Here, those skilled in the art will appreciate that the specific operations of each module in the above nursing data management optimization system for brain tumor patients have been described in the above Figures 1 to 5 The present invention has been described in detail in the description of the nursing data management optimization method for brain tumor patients, and therefore, its repeated description will be omitted.

[0095] 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 nursing data management optimization method for brain tumor patients, characterized in that: include: Acquiring vital sign data of a target brain tumor patient subject, and performing FHIR conversion on the vital sign data of the target brain tumor patient subject to obtain vital sign FHIR format data; obtaining electronic medical record data of the target brain tumor patient; Extract key data from the electronic medical record data and perform FHIR conversion using an FHIR adapter to obtain electronic medical record FHIR format data; Acquiring medical imaging data of the target brain tumor patient; Performing a medical image structured description based on a large model on the medical image data of the target brain tumor patient to obtain a medical image structured description, and performing FHIR conversion on the medical image structured description to obtain medical image structured description FHIR format data; Deploy a server that complies with the FHIR standard, wherein the server that complies with the FHIR standard is used to store and manage the vital sign FHIR format data, the electronic medical record FHIR format data, and the medical image structured description FHIR format data; The FHIR-compliant server provides a standard FHIR API for other application systems to access and query data.

2. The nursing data management optimization method for brain tumor patients according to claim 1 is characterized in that: The electronic medical record data is extracted and converted to FHIR format using the FHIR adapter to obtain electronic medical record FHIR format data, including: extracting key data elements from the electronic medical record data to obtain a set of key data elements; Based on the data mapping rule, mapping each key data element in the set of key data elements to key data element FHIR format data to obtain a set of key data element FHIR format data; Data verification and quality control are performed on the set of key data element FHIR format data to obtain the electronic medical record FHIR format data.

3. The nursing data management optimization method for brain tumor patients according to claim 2 is characterized in that: The data mapping rules include mapping of field names, conversion of data types, conversion of encoding systems, and conversion of data values.

4. The nursing data management optimization method for brain tumor patients according to claim 1 is characterized in that: Performing a large-model-based medical image structured description on the medical image data of the target brain tumor patient to obtain a medical image structured description, including: extracting lesion visual features from medical imaging data of the target brain tumor patient to obtain a set of lesion local visual feature coding maps; Performing visual feature space saliency on each lesion local visual feature coding map in the set of lesion local visual feature coding maps to obtain a set of lesion local visual feature saliency coding maps; fusing the set of significant coding maps of local visual features of the lesion to obtain an aggregated coding map of local visual features of the lesion; Performing image semantic segmentation on the lesion local visual feature aggregation coding map to obtain an image semantic segmentation result; The image semantic segmentation result is represented in a structured format to obtain the medical image structured description.

5. The nursing data management optimization method for brain tumor patients according to claim 4 is characterized in that: Extracting lesion visual features from medical imaging data of the target brain tumor patient to obtain a set of lesion local visual feature coding maps, including: performing image denoising on the medical image data of the target brain tumor patient to obtain denoised medical image data of the target brain tumor patient; A lesion visual feature extractor based on a deep separable convolutional neural network is used to extract local image features from the medical imaging data of the target brain tumor patient after noise reduction to obtain a set of lesion local visual feature coding maps.

6. The nursing data management optimization method for brain tumor patients according to claim 5, characterized in that: Performing visual feature space saliency on each lesion local visual feature coding map in the set of lesion local visual feature coding maps to obtain a set of lesion local visual feature saliency coding maps, comprising: Extracting the channel feature vector of the (i, j)th pixel position from the lesion local visual feature coding map as the lesion local pixel-level visual feature vector to be enhanced; Based on the feature distribution of the local pixel-level visual feature vector to be enhanced of the lesion, anchoring the local receptive field of the lesion local visual feature encoding map to screen out a set of local pixel-level visual feature vectors of the lesion within the local receptive field; Based on the set of local pixel-level visual feature vectors of the lesion within the local receptive field, feature saliency is performed on the local pixel-level visual feature vectors of the lesion to be enhanced based on context feature association guidance to obtain a local pixel-level visual feature saliency coding vector of the lesion, wherein the local pixel-level visual feature saliency coding vector of the lesion is the channel feature vector of the (i, j)th pixel position in the local visual feature saliency coding map of the lesion.

7. The nursing data management optimization method for brain tumor patients according to claim 6 is characterized in that: Based on the feature distribution of the local pixel-level visual feature vector to be enhanced of the lesion, local receptive field anchoring is performed on the local visual feature encoding map of the lesion to screen out a set of local pixel-level visual feature vectors of the lesion within the local receptive field, including: Performing feature decoupling on the lesion local feature map along the channel dimension to obtain a set of lesion local pixel-level visual feature vectors; Compressing the pixel-level visual feature vector of the lesion to be enhanced to obtain a compressed coding vector of the pixel-level visual feature of the lesion to be enhanced; Determining the size of the feature receptive field of the pixel-level visual feature compression coding vector of the local lesion to be enhanced based on the feature distribution spatial structure characteristics of the pixel-level visual feature compression coding vector of the local lesion to be enhanced; Based on the size of the feature receptive field, a set of local pixel-level visual feature vectors of the lesion within the local receptive field is filtered out from the set of local pixel-level visual feature vectors of the lesion.

8. The nursing data management optimization method for brain tumor patients according to claim 7, characterized in that: Based on a set of local pixel-level visual feature vectors of the lesion within the local receptive field, performing feature saliency on the local pixel-level visual feature vectors of the lesion to be enhanced based on context feature association guidance to obtain a salient encoding vector of the local pixel-level visual feature of the lesion, including: Based on the feature correlation between each local pixel-level visual feature vector of the lesion in the local receptive field and the local pixel-level visual feature vector of the lesion to be enhanced, the set of local pixel-level visual feature vectors of the lesion in the local receptive field is subjected to saliency aggregation to obtain a saliency aggregation coding vector of the local pixel-level visual feature of the lesion in the local receptive field; The significant aggregation coding vector of the local pixel-level visual feature of the lesion in the local receptive field and the local pixel-level visual feature vector to be enhanced of the lesion are fused to obtain the significant coding vector of the local pixel-level visual feature of the lesion.

9. The nursing data management optimization method for brain tumor patients according to claim 8, characterized in that: The saliency aggregation coding vector of the local pixel-level visual feature of the lesion in the local receptive field and the local pixel-level visual feature vector of the lesion to be enhanced are fused to obtain the saliency coding vector of the local pixel-level visual feature of the lesion, comprising: Based on the conformal fusion regularity constraint of the set of local pixel-level visual feature vectors of the lesion in the local receptive field, modulating the weighting coefficients of the saliency aggregation coding vector of the local pixel-level visual feature of the lesion in the local receptive field and the local pixel-level visual feature vector to be enhanced of the lesion; Based on the weighting coefficient, the lesion local pixel-level visual feature saliency aggregation coding vector within the local receptive field and the lesion local pixel-level visual feature vector to be enhanced are weightedly fused to obtain the lesion local pixel-level visual feature saliency coding vector.

10. A nursing data management optimization system for brain tumor patients, characterized in that: include: a vital sign data FHIR conversion module, configured to obtain vital sign data of a target brain tumor patient object, and perform FHIR conversion on the vital sign data of the target brain tumor patient object to obtain vital sign FHIR format data; an electronic medical record data acquisition module, configured to acquire electronic medical record data of the target brain tumor patient; An electronic medical record FHIR conversion module is used to extract key data and perform FHIR conversion on the electronic medical record data through an FHIR adapter to obtain electronic medical record FHIR format data; a medical imaging data acquisition module, configured to acquire medical imaging data of the target brain tumor patient; a medical image structured description module, configured to perform a medical image structured description based on a large model on the medical image data of the target brain tumor patient to obtain a medical image structured description, and perform FHIR conversion on the medical image structured description to obtain medical image structured description FHIR format data; A server deployment module is used to deploy a server that complies with the FHIR standard. The server that complies with the FHIR standard is used to store and manage the vital signs FHIR format data, the electronic medical record FHIR format data, and the medical image structured description FHIR format data. The server that complies with the FHIR standard provides a standard FHIR API for other application systems to access and query data.

Images