Intelligent triage system, method and equipment based on large medical model and medium

Abstract

The invention discloses an intelligent triage system, method and device based on a medical large model and a medium, and relates to the technical field of medical diagnosis, and the system comprises a first data obtaining interface which is used for obtaining multi-modal medical record data; the data fusion tool is used for analyzing the multi-modal medical record data and fusing the multi-modal medical record data according to an analysis result to obtain fused medical record data; the second data acquisition interface is used for acquiring real-time influenza data, regional climate information and target operation data; the data input module is used for inputting the fused medical record data, the real-time influenza data, the regional climate information and the target operation data into a target medical big model, so that the target medical big model generates target triage information corresponding to each target patient; and the information acquisition module is used for acquiring target triage information. The patient is triage by using the influenza data, the regional climate information and the operation data of the hospital, so that the reliability of the triage result is ensured.

Description

Technical Field

[0001] The present invention relates to the field of medical diagnosis technology, and in particular to an intelligent triage system, method, device and medium based on a large medical model. Background Art

[0002] Triage involves the rapid and focused collection of patient data, analysis, assessment, classification, and departmentalization, while also arranging visits based on severity, urgency, and non-urgent conditions. Triage effectively controls the number of patients in the emergency room, maintains order within the room, arranges appropriate treatment locations, and increases patient satisfaction with the diagnosis.

[0003] Currently, triage relies primarily on electronic tools based on rule engines, such as those that match symptoms with keywords and use automated departmental systems. This approach fails to dynamically integrate with real-time medical data (such as infectious disease monitoring, departmental patient load, and drug inventory), resulting in a disconnect between triage recommendations and the actual state of medical resources. Therefore, improving the reliability of triage results has become a pressing technical challenge in the patient triage process. Summary of the Invention

[0004] In light of this, the present invention aims to provide an intelligent triage system, method, device, and medium based on a large medical model. By utilizing influenza data, regional climate information, and hospital operational data to triage patients, the system ensures the reliability of triage results. The specific solution is as follows:

[0005] In a first aspect, the present application provides an intelligent triage system based on a large medical model, applied to a computer device, comprising:

[0006] The first data acquisition interface is used to obtain multimodal medical record data corresponding to the target patient;

[0007] A data fusion tool for analyzing each of the multimodal medical record data according to the data type corresponding to each of the multimodal medical record data, and fusing each of the multimodal medical record data according to the corresponding analysis results to obtain fused medical record data;

[0008] A second data acquisition interface is used to obtain real-time influenza data, regional climate information, and target operational data using a preset application program interface; wherein the target operational data includes the patient load of each department, medical staff attendance, and drug inventory status;

[0009] a data input module, configured to input the fused medical record data, the real-time influenza data, the regional climate information, and the target operational data into a target medical macromodel, so that the target medical macromodel generates target triage information corresponding to each target patient based on the fused medical record data, the real-time influenza data, the regional climate information, and the target operational data; wherein the target medical information includes waiting time, medical department, and medical advice;

[0010] The information acquisition module is used to obtain the target triage information corresponding to each target patient output by the target medical model.

[0011] Optionally, the data fusion tool includes:

[0012] a speech recognition unit, configured to recognize speech data in the multimodal medical record data based on a preset dialect phoneme library and a preset medical term dictionary, so as to map non-standardized expressions in the speech data into corresponding clinical medical terms; wherein the preset dialect phoneme library is configured to store phonemes corresponding to different dialects, and the preset medical term dictionary is configured to record clinical medical terms;

[0013] A text analysis unit, configured to train a preset initial BERT model using a domain adaptation strategy to obtain a trained BERT model, and analyze text data in the multimodal medical record data using the trained BERT model and natural language processing technology to obtain text features corresponding to the text data;

[0014] an image normalization unit, configured to determine image data in the multimodal medical record data and perform normalization processing on each image data to obtain a normalized image corresponding to each image data;

[0015] A data fusion module is used to fuse the multimodal data based on the clinical medical terms, the text features and the normalized image corresponding to the voice data.

[0016] Optionally, the data fusion module includes:

[0017] a cosine similarity determination unit, configured to determine, based on the clinical medical term, the text feature, and the normalized image, a first cosine similarity between the voice data and the text data, a second cosine similarity between the voice data and the image data, and a third cosine similarity between the text data and the image data, and determine whether the first cosine similarity, the second cosine similarity, and the third cosine similarity are less than a preset similarity threshold;

[0018] a data fusion unit, configured to dynamically set first target weights corresponding to the voice data, the text features, and the image data, respectively, according to the disease type corresponding to the multimodal medical record data, if the first cosine similarity, the second cosine similarity, and the third cosine similarity are all not less than the preset similarity threshold, and fuse the voice data, the text features, and the image data according to the first target weights.

[0019] Optionally, the target medical model includes:

[0020] a weight adjustment unit, configured to predict seasonal influenza trends based on the real-time influenza data and the regional climate information, dynamically adjust second target weights corresponding to different departments based on the seasonal influenza trends, and determine target matching degrees between the target patient and different departments using the second target weights and the fused medical record data;

[0021] A load forecasting unit, configured to forecast the patient load of each department within a preset time period based on the target operation data to obtain corresponding forecast results;

[0022] The first medical information generating unit is used to generate the target triage information corresponding to each target patient based on the target matching degree, the prediction result and the second target weights corresponding to different departments.

[0023] Optionally, the target medical model includes:

[0024] The second medical information generating unit is used to determine the Pareto optimal solution corresponding to each target patient by using a multi-objective optimization algorithm, and generate the target triage information corresponding to each target patient according to the Pareto optimal solution.

[0025] Optionally, the information acquisition module further includes:

[0026] An information adjustment unit is used to determine whether the confidence of the target triage information is less than a preset confidence threshold. If the confidence of the target triage information is less than the preset confidence threshold, the target triage information is adjusted to obtain corresponding adjusted triage information.

[0027] Optionally, the intelligent triage system further includes:

[0028] The model training module is used to iteratively train the target medical model using the target training set and the elastic weight consolidation algorithm to update the model parameters of the target medical model if the confidence of the target triage information is less than the preset confidence threshold.

[0029] In a second aspect, the present application provides an intelligent triage method based on a large medical model, applied to a computer device, comprising:

[0030] Obtain multimodal medical record data corresponding to the target patient;

[0031] Analyzing each of the multimodal medical record data according to the data type corresponding to each of the multimodal medical record data, and fusing each of the multimodal medical record data according to the corresponding analysis results to obtain fused medical record data;

[0032] Utilize pre-defined application programming interfaces to obtain real-time influenza data, regional climate information, and target operational data, including department load, medical staff attendance, and drug inventory status.

[0033] Inputting the fused medical record data, the real-time influenza data, the regional climate information, and the target operation data into a target medical big model, so that the target medical big model generates target triage information corresponding to each target patient based on the fused medical record data, the real-time influenza data, the regional climate information, and the target operation data; wherein the target medical information includes waiting time, medical department, and medical advice;

[0034] The target triage information corresponding to each target patient output by the target medical model is obtained.

[0035] In a third aspect, the present application provides an electronic device comprising a processor and a memory; wherein the memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the aforementioned intelligent triage method based on the medical big model.

[0036] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the aforementioned intelligent triage method based on a large medical model.

[0037] The intelligent triage system in this application includes: a first data acquisition interface for acquiring multimodal medical record data corresponding to a target patient; a data fusion tool for analyzing each multimodal medical record data according to the data type corresponding to each multimodal medical record data, and fusing each multimodal medical record data according to the corresponding analysis results to obtain fused medical record data; a second data acquisition interface for acquiring real-time influenza data, regional climate information and target operation data using a preset application program interface; wherein the target operation data includes the reception load of each department, the attendance of medical staff and the inventory of medicines storage status; a data input module, used to input the fused medical record data, the real-time influenza data, the regional climate information and the target operation data into the target medical big model, so that the target medical big model generates target triage information corresponding to each target patient according to the fused medical record data, the real-time influenza data, the regional climate information and the target operation data; wherein the target medical information includes waiting time, medical department and medical advice; an information acquisition module, used to obtain the target triage information corresponding to each target patient output by the target medical big model. It can be seen that this application uses a preset application interface to obtain real-time influenza data, regional climate information and hospital operation data, and triages patients based on influenza data, regional climate information and hospital operation data, thereby dynamically linking the triage process with the implemented medical resources to ensure the reliability of the triage results; by obtaining the patient's multimodal medical record information, fusing the multimodal medical record information and performing triage based on the fused case information, it avoids triaging patients based only on a single modality of data, thereby improving the reliability of the triage results. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0039] Figure 1 A schematic diagram of the structure of an intelligent triage system based on a large medical model provided in this application;

[0040] Figure 2 A flowchart of an intelligent triage method based on a large medical model provided in this application;

[0041] Figure 3 A flowchart of an intelligent triage method based on a large medical model provided in this application;

[0042] Figure 4 This is a structural diagram of an electronic device provided in this application. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] Currently, patients are mainly triaged using electronic tools based on rule engines. This method has the problem of triage recommendations being out of touch with the actual status of medical resources. To this end, this application provides an intelligent triage system based on a large medical model. By utilizing influenza data, regional climate information, and hospital operational data to triage patients, the reliability of the triage results is ensured.

[0045] See also Figure 1 As shown, the embodiment of the present invention discloses an intelligent triage system based on a medical large model, which is applied to a computer device and includes a first data acquisition interface 11, a data fusion tool 12, a second data acquisition interface 13, a data input module 14 and an information acquisition module 15; wherein,

[0046] The first data acquisition interface 11 is used to acquire multimodal medical record data corresponding to a target patient.

[0047] This embodiment aims to build a multimodal, dynamically perceived, and securely controllable intelligent triage system. Its core objectives include: overcoming the limitations of single-modal data processing by integrating natural language understanding, medical image analysis, and voice interaction technologies to achieve a three-dimensional analysis of patient symptoms; establishing a real-time connection mechanism with public health warning systems and hospital operational data to adapt triage decisions to dynamic environmental factors such as seasonal epidemic changes and departmental capacity; designing a traceable decision path visualization solution to enhance medical staff's trust in AI recommendations through features such as a symptom-department matching matrix and similar case comparisons; employing a federated learning framework to achieve cross-institutional collaborative training, enhancing model generalization while ensuring patient privacy; and ultimately forming a complete closed loop of "data collection - intelligent analysis - decision output - feedback optimization," driving the transformation of the triage process from experience-driven to data-intelligent-driven.

[0048] In this embodiment, the overall workflow of the intelligent triage system is as follows: Figure 2 As shown, it is first necessary to obtain the patient's corresponding multimodal medical record data such as voice, text, and images, then analyze the multimodal data and output the analysis results. If necessary, the output results can also be fed back and optimized.

[0049] It should be noted that this system adopts a multi-layer fusion architecture, with a medical big model as the core, to build a full-process technology system covering data perception, intelligent decision-making, human-computer interaction and privacy protection. The system integrates patient complaints, medical images, electronic health records and IoT device monitoring data through a multimodal data acquisition module, and uses natural language processing, computer vision and time series data analysis technology to achieve a three-dimensional analysis of symptoms. At the core layer, the medical big model based on the Transformer architecture (a deep learning model architecture) completes knowledge accumulation and scenario adaptation through a two-stage training strategy: the pre-training stage uses a comparative learning strategy to align massive medical literature and clinical guidelines to form a semantic association network of symptoms, diseases and departments; the fine-tuning stage introduces a reinforcement learning mechanism, using the final diagnosis result as a delayed reward signal to optimize the triage decision logic.

[0050] The data fusion tool 12 is used to analyze each of the multimodal medical record data according to the data type corresponding to each of the multimodal medical record data, and to fuse each of the multimodal medical record data according to the corresponding analysis results to obtain fused medical record data.

[0051] In this embodiment, the data fusion tool 12 includes: a speech recognition unit for recognizing speech data in the multimodal medical record data based on a preset dialect phoneme library and a preset medical term dictionary, so as to map non-standardized expressions in the speech data into corresponding clinical medical terms; wherein the preset dialect phoneme library is used to store phonemes corresponding to different dialects, and the preset medical term dictionary is used to record clinical medical terms;

[0052] A text analysis unit is used to train a preset initial BERT (Bidirectional Encoder Representations from Transformers) model using a domain adaptation strategy to obtain a trained BERT model, and then use the trained BERT model and natural language processing technology to analyze the text data in the multimodal medical record data to obtain text features corresponding to the text data;

[0053] An image normalization unit, configured to determine image data in the multimodal medical record data and perform normalization processing on each image data to obtain a normalized image corresponding to each image data;

[0054] The data fusion module is used to fuse multimodal data based on clinical medical terms, text features and normalized images corresponding to the speech data.

[0055] In this embodiment, the data fusion module 12 may specifically include: a cosine similarity determination unit, configured to determine, based on clinical medical terms, text features, and normalized images, a first cosine similarity between the voice data and the text data, a second cosine similarity between the voice data and the image data, and a third cosine similarity between the text data and the image data, and determine whether the first cosine similarity, the second cosine similarity, and the third cosine similarity are less than a preset similarity threshold;

[0056] The data fusion unit is configured to dynamically set first target weights corresponding to the voice data, text features, and image data, respectively, based on the disease type corresponding to the multimodal medical record data, if the first cosine similarity, the second cosine similarity, and the third cosine similarity are all no less than a preset similarity threshold, and to fuse the voice data, text features, and image data according to each first target weight. It will be appreciated that if the cosine similarity between any two of them is less than the preset similarity threshold, this indicates an anomaly in the matching of the multimodal data, requiring the multimodal data to be reloaded and the cosine similarity thresholds between the modal data to be recalculated after review.

[0057] Specifically, the system's multimodal perception layer establishes an efficient processing channel for heterogeneous data, enabling collaborative analysis of voice, text, and images. The voice interaction module, based on the Wav2Vec 2.0 architecture (a pre-trained model), has been optimized for medical scenarios. It integrates a pre-set dialect phoneme library and a dictionary of specialized terminology, enabling accurate recognition of ambiguous symptom descriptions with regional accents (e.g., parsing "heart pain" as "angina pectoris"). The text parsing unit utilizes a domain-adaptive training strategy and builds a medical entity recognition system based on the BERT model. This system covers 12 categories of clinical entities (i.e., text features), including symptoms, signs, medical history, and allergens. Furthermore, to address the temporal nature of symptom descriptions, the system incorporates a Bidirectional Long Short-Term Memory (BiLSTM) network to capture key temporal logic, such as "rash appears three days after fever." This is combined with a timestamp encoder to construct a symptom evolution trajectory. This allows the system to construct a symptom evolution trajectory based on the patient's textual descriptions and the timestamp encoder, enhancing its understanding of the patient's condition. The medical image analysis module features a standardized preprocessing pipeline that supports automatic conversion and normalization of multiple formats, including DICOM (Digital Imaging and Communications in Medicine) and JPG (an image format). The lesion detection model utilizes an improved U-Net (a neural network) architecture, embedding a spatial attention mechanism within residual connections to effectively enhance the ability to identify small lesions. The multimodal alignment verification mechanism uses comparative learning to verify image-text consistency. For example, when a patient describes an "annular erythema," the system automatically calculates the cosine similarity between the dermatoscope image and the text features. If the cosine similarity falls below a preset threshold (i.e., the preset similarity threshold) of 0.75, a manual review process is triggered. The data fusion layer adopts a gated multi-head attention mechanism to dynamically adjust the contribution weights of different modal features. For example, in the acute chest pain scenario, the weight of the ECG monitoring data (i.e., the first target weight) is increased to 0.6, while the weight of the speech description (the first target weight) is correspondingly reduced to 0.2, ensuring that key vital sign data dominates the decision-making process. In other words, this system can dynamically adjust the weight of data fusion in combination with different disease scenarios, thereby ensuring a high degree of consistency between the fused data and the disease.

[0058] The second data acquisition interface 13 is used to obtain real-time influenza data, regional climate information and target operation data using a preset application program interface; wherein the target operation data includes the patient load of each department, the attendance of medical staff and the inventory status of medicines.

[0059] In this embodiment, the dynamic knowledge engine in the system accesses public health warning data and hospital operation status in real time, and embeds environmental parameters such as department load rate and drug inventory into the decision model through a spatiotemporal feature encoder to ensure dynamic adaptation of triage recommendations to the real-time status of medical resources. That is, the dynamic knowledge engine in this embodiment includes: a static knowledge base for integrating SNOMED CT (Systematized Nomenclature of Medicine -- Clinical Terms) and ICD-11 (International Classification of Diseases) to construct a symptom, disease, and department triplet map; a dynamic data stream processor for real-time access to epidemic API (Application Programming Interface) and HIS (Hospital Information System) system operation data; and a spatiotemporal encoder for fusing department load rate and seasonal epidemic characteristics using a GRU network.

[0060] The core competitiveness of the medical big model in this embodiment stems from its integrated static and dynamic knowledge management system. The static knowledge base integrates authoritative guidelines such as the SNOMED CT clinical terminology system, ICD-11 disease classification codes, and UpToDate, constructing a symptom-disease-department triple graph containing 5.3 million entity relationships. The dynamic data stream processor accesses epidemic monitoring APIs, HIS system operational data (i.e., target operational data), and regional climate information in real time.

[0061] The data input module 14 is configured to input the fused medical record data, the real-time influenza data, the regional climate information, and the target operational data into the target medical model, so that the target medical model generates target triage information corresponding to each target patient based on the fused medical record data, the real-time influenza data, the regional climate information, and the target operational data. The target triage information includes waiting time, department, and recommended treatment.

[0062] In this example, the target medical big model includes: a weight adjustment unit, which is used to predict the seasonal influenza trend based on the real-time influenza data and the regional climate information, dynamically adjust the second target weights corresponding to different departments according to the seasonal influenza trend, and use the second target weights and the fused medical record data to determine the target matching degree between the target patients and different departments; a load forecasting unit, which is used to forecast the reception load of each department within a preset time period based on the target operation data to obtain the corresponding forecast results; a first consultation information generation unit, which is used to generate target triage information corresponding to each target patient based on the target matching degree, the forecast results and the second target weights corresponding to different departments. Specifically, a time series prediction model (such as the Prophet algorithm) is used to predict the seasonal epidemic trend. When the influenza warning level increases, the system automatically adjusts the triage strategy, increases the respiratory department weight coefficient (i.e., the second target weight) by 30%, and adds relevant examination item recommendations. The department load perception module obtains the real-time number of patients, doctor attendance status, and equipment availability data of each department through a distributed message queue (Apache Kafka), and uses the exponential smoothing method to predict the reception pressure in the next two hours, providing a quantitative basis for dynamic resource allocation.

[0063] The construction of a three-dimensional decision-making model that integrates department load, epidemiological trends, and the match between symptoms and departments embodies a deep integration of clinical logic and operational optimization. In the professional dimension, the system calculates the cosine similarity between symptom vectors and department vectors to quantify the medical relevance between symptoms and departments. For example, the symptom "chest pain radiating to the left arm" has a match of 0.91 with cardiovascular medicine, but only 0.35 with gastroenterology. The resource dimension introduces a dynamic decay function to adjust recommendation priorities based on the real-time load of departments: when the number of patients waiting in a department reaches 80% of its maximum capacity, its weight coefficient decreases exponentially, prompting the system to prioritize alternative departments with lower loads. The personalized adaptation dimension integrates personalized factors such as patient insurance type, allergy history, and transportation accessibility. For example, it can automatically screen for lower-cost examination options for patients with insufficient medical insurance coverage or recommend nearby medical centers for those with limited mobility.

[0064] In addition, the decision model in this embodiment uses a three-dimensional optimization algorithm to balance clinical priority and resource utilization, and constructs a multi-dimensional decision space based on symptom professional fit, department reception capabilities, and individual patient characteristics. For emergency scenarios, the system improves the traditional MEWS (Modified Early Warning Score) scoring model and extracts non-contact vital sign data such as respiratory rate through video analysis technology to achieve rapid identification of critical and severe illnesses and prioritize resource allocation. The human-computer collaborative interface designs an explainable decision panel, using LRP (Layerwise Relevance Propagation, an algorithm for interpreting deep neural networks) technology to visualize the symptom weight distribution and associate the complete diagnosis and treatment paths of similar historical cases, providing medical staff with a transparent decision-making basis. The privacy protection system runs through the entire data life cycle, realizes cross-institutional model collaborative training through a federated learning framework, adopts differential privacy and homomorphic encryption technologies to ensure the secure flow of sensitive information, and uses generative adversarial networks to de-identify medical images, meeting medical data compliance requirements while ensuring diagnostic accuracy. That is, the human-computer collaborative interface in this embodiment includes: a three-dimensional heat map visualization module for presenting the probability distribution of department recommendations in color gradients; a case tracing system for comparing diagnosis and treatment pathways of similar historical cases; an explainable analysis unit for highlighting the decision-making basis characteristics using the LIME (Local Interpretable Model-agnostic Explanations, an algorithm for improving the interpretability of machine learning models) algorithm; through multimodal data fusion and three-dimensional decision models, the completeness of symptom analysis and the accuracy of department matching are significantly improved; and by integrating real-time vital sign monitoring with an improved scoring algorithm, the identification and treatment time of critically ill patients is shortened.

[0065] In this embodiment, the target medical macromodel includes a second medical information generation unit, which utilizes a multi-objective optimization algorithm to determine the Pareto optimal solution for each target patient and, based on this Pareto optimal solution, generates target triage information for each target patient. Specifically, the NSGA-III (Non-dominated Sorting Genetic Algorithms) algorithm is used to search for the Pareto optimal solution within this three-dimensional space, balancing the conflicting requirements of department matching, wait time, and treatment cost. Ultimately, a structured report is generated, including a confidence score, recommended examination items, and referral recommendations.

[0066] The information acquisition module 15 is used to obtain the target triage information corresponding to each target patient output by the target medical model.

[0067] In this embodiment, the information acquisition module 15 further includes: an information adjustment unit for determining whether the confidence of the target triage information is less than a preset confidence threshold. If the confidence of the target triage information is less than the preset confidence threshold, the target triage information is adjusted to obtain the corresponding adjusted triage information. Specifically, the system is provided with a feedback optimization closed-loop design dual-channel learning mechanism: when the system recommendation confidence is lower than the preset confidence threshold, for example, 85%, a manual review process is triggered, and medical staff can choose from three alternative options or manually modify the department. The labeled data reviewed by two deputy chief physicians enters the incremental training queue.

[0068] In this embodiment, the intelligent triage system also includes: a model training module, which is used to iteratively train the target medical model using the target training set and the elastic weight consolidation algorithm if the confidence of the target triage information is less than the preset confidence threshold, so as to update the model parameters of the target medical model. Specifically, the model uses the elastic weight consolidation EWC (ElasticWeight Consolidation) algorithm for online learning, and retains important parameters by calculating the Fisher information matrix to prevent new knowledge from covering existing experience. The knowledge base version management implements strict two-tier control: routine updates are automatically completed by the system, and operations involving major revisions to the diagnosis and treatment guidelines must be approved by the Medical Ethics Committee to ensure the scientificity and security of knowledge evolution.

[0069] In addition, the system also has a human-computer collaborative interface, which builds a symbiotic relationship between clinical experience and AI (Artificial Intelligence) models through visualization and feedback mechanisms. The decision panel uses a three-dimensional heat map to present the recommended distribution of departments. The dark red area represents the core department with a matching degree higher than 90%. Medical staff can rotate and observe the multi-angle decision space through gestures. The case tracing system links to a library of similar historical cases to show the differences between the current patient and the typical case in key feature dimensions, such as "white blood cell count vs. Average A significant abnormality in the "" will trigger a sepsis warning. The explainability analysis module uses the Local Interpretable Model (LIME) technology to highlight the core factors that influence decision-making. For example, "pregnancy status" accounts for 41% of the weight of triage recommendations, and the system automatically blocks radiology examination recommendations based on this.

[0070] It should be noted that the system employs a privacy protection system throughout the entire process of data collection, transmission, storage, and analysis. The federated learning framework employs a hybrid horizontal and vertical architecture: horizontal federation enables 10 hospital nodes to collaboratively train models without sharing raw data, aggregating gradient parameters through Paillier homomorphic encryption technology; vertical federation enables cross-departmental feature space alignment, such as securely correlating blood indicators from the laboratory department with CT features from the imaging department. The data desensitization module utilizes a generative adversarial network (CycleGAN) to synthesize anonymous medical images, completely eliminating patient identification information while preserving diagnostic value. The access control system implements refined management based on RBAC (Role-Based Access Control). All data operation records are stored on the blockchain (Hyperledger Fabric) to ensure that operation traces cannot be tampered with.

[0071] The system utilizes a collaborative cloud-edge-device deployment model. NVIDIA Jetson AGX Xavier devices are deployed at the edge to achieve real-time clinic-level response, keeping inference latency under 500ms. A Kubernetes containerized cluster is built on the cloud to manage model training and knowledge graph updates, and cross-node data synchronization is achieved via a high-speed RDMA (Remote Direct Memory Access) network. The disaster recovery system utilizes active-active data centers and Ceph distributed storage to ensure business continuity, achieving an RPO (Recovery Point Objective) of less than 15 seconds. In extreme situations such as network outages, the system automatically switches to a local knowledge base snapshot mode to maintain basic triage functions.

[0072] It can be seen that this application uses a preset application interface to obtain real-time influenza data, regional climate information and hospital operation data, and triages patients based on influenza data, regional climate information and hospital operation data, thereby dynamically linking the triage process with the implemented medical resources to ensure the reliability of the triage results; by obtaining the patient's multimodal medical record information, fusing the multimodal medical record information and performing triage based on the fused case information, it avoids triaging patients based only on a single modality of data, thereby improving the reliability of the triage results.

[0073] Based on the above embodiments, it can be seen that this application discloses an intelligent triage system based on a large medical model, which can use influenza data, regional climate information and hospital operation data to triage patients, ensuring the reliability of the triage results. Next, this embodiment will explain the corresponding intelligent triage method. Figure 3As shown, the embodiment of the present application also discloses an intelligent triage method based on a large medical model, which is applied to a computer device, including:

[0074] Step S11: Obtain multimodal medical record data corresponding to the target patient.

[0075] Step S12: Analyze each of the multimodal medical record data according to the data type corresponding to each of the multimodal medical record data, and fuse each of the multimodal medical record data according to the corresponding analysis results to obtain fused medical record data.

[0076] Step S13: Utilize a preset application program interface to obtain real-time influenza data, regional climate information, and target operation data; wherein the target operation data includes the patient load of each department, medical staff attendance, and drug inventory status.

[0077] Step S14: input the fused medical record data, the real-time influenza data, the regional climate information and the target operation data into the target medical big model, so that the target medical big model generates target triage information corresponding to each target patient according to the fused medical record data, the real-time influenza data, the regional climate information and the target operation data; wherein the target medical information includes waiting time, medical department and medical advice.

[0078] Step S15: Obtain the target triage information corresponding to each target patient output by the target medical model.

[0079] The above-mentioned intelligent triage method based on the medical big model has the same beneficial effects as the aforementioned intelligent triage system based on the medical big model, and will not be described in detail here.

[0080] It can be seen that this application uses a preset application interface to obtain real-time influenza data, regional climate information and hospital operation data, and triages patients based on influenza data, regional climate information and hospital operation data, thereby dynamically linking the triage process with the implemented medical resources to ensure the reliability of the triage results; by obtaining the patient's multimodal medical record information, fusing the multimodal medical record information and performing triage based on the fused case information, it avoids triaging patients based only on a single modality of data, thereby improving the reliability of the triage results.

[0081] Furthermore, the embodiment of the present application also discloses an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram should not be considered as any limitation to the scope of application of the present application.

[0082] Figure 4This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps of the intelligent triage method based on a large medical model disclosed in any of the aforementioned embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0083] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0084] In addition, the memory 22 as a carrier for resource storage can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon may include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0085] The operating system 221 is used to manage and control the hardware devices on the electronic device 20 and the computer program 222. The operating system 221 can be Windows Server, NetWare, Unix, Linux, etc. In addition to including computer programs capable of implementing the intelligent triage method based on the medical big model and executed by the electronic device 20 as disclosed in any of the aforementioned embodiments, the computer program 222 can further include computer programs capable of performing other specific tasks.

[0086] Furthermore, this application discloses a computer-readable storage medium for storing a computer program; wherein, when executed by a processor, the computer program implements the aforementioned intelligent triage method based on a large medical model. The specific steps of this method can be found in the corresponding contents disclosed in the aforementioned embodiments and will not be repeated here.

[0087] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0088] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0089] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0090] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0091] The above is a detailed introduction to the technical solution provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. An intelligent triage system based on a large medical model, characterized by: Applicable to computer devices, including: The first data acquisition interface is used to obtain multimodal medical record data corresponding to the target patient; A data fusion tool for analyzing each of the multimodal medical record data according to the data type corresponding to each of the multimodal medical record data, and fusing each of the multimodal medical record data according to the corresponding analysis results to obtain fused medical record data; A second data acquisition interface is used to obtain real-time influenza data, regional climate information, and target operational data using a preset application program interface; wherein the target operational data includes the patient load of each department, medical staff attendance, and drug inventory status; a data input module, configured to input the fused medical record data, the real-time influenza data, the regional climate information, and the target operational data into a target medical macromodel, so that the target medical macromodel generates target triage information corresponding to each target patient based on the fused medical record data, the real-time influenza data, the regional climate information, and the target operational data; wherein the target medical information includes waiting time, medical department, and medical advice; The information acquisition module is used to obtain the target triage information corresponding to each target patient output by the target medical model.

2. The intelligent triage system based on the medical big model according to claim 1 is characterized in that: The data fusion tool includes: a speech recognition unit, configured to recognize speech data in the multimodal medical record data based on a preset dialect phoneme library and a preset medical term dictionary, so as to map non-standardized expressions in the speech data into corresponding clinical medical terms; wherein the preset dialect phoneme library is configured to store phonemes corresponding to different dialects, and the preset medical term dictionary is configured to record clinical medical terms; A text analysis unit, configured to train a preset initial BERT model using a domain adaptation strategy to obtain a trained BERT model, and analyze text data in the multimodal medical record data using the trained BERT model and natural language processing technology to obtain text features corresponding to the text data; an image normalization unit, configured to determine image data in the multimodal medical record data and perform normalization processing on each image data to obtain a normalized image corresponding to each image data; A data fusion module is used to fuse the multimodal data based on the clinical medical terms, the text features and the normalized image corresponding to the voice data.

3. The intelligent triage system based on the medical big model according to claim 2 is characterized in that: The data fusion module includes: a cosine similarity determination unit, configured to determine, based on the clinical medical term, the text feature, and the normalized image, a first cosine similarity between the voice data and the text data, a second cosine similarity between the voice data and the image data, and a third cosine similarity between the text data and the image data, and determine whether the first cosine similarity, the second cosine similarity, and the third cosine similarity are less than a preset similarity threshold; a data fusion unit, configured to dynamically set first target weights corresponding to the voice data, the text features, and the image data, respectively, according to the disease type corresponding to the multimodal medical record data, if the first cosine similarity, the second cosine similarity, and the third cosine similarity are all not less than the preset similarity threshold, and fuse the voice data, the text features, and the image data according to the first target weights.

4. The intelligent triage system based on the medical big model according to claim 1 is characterized in that: The target medical model includes: a weight adjustment unit, configured to predict seasonal influenza trends based on the real-time influenza data and the regional climate information, dynamically adjust second target weights corresponding to different departments based on the seasonal influenza trends, and determine target matching degrees between the target patient and different departments using the second target weights and the fused medical record data; A load forecasting unit, configured to forecast the patient load of each department within a preset time period based on the target operation data to obtain corresponding forecast results; The first medical information generating unit is used to generate the target triage information corresponding to each target patient based on the target matching degree, the prediction result and the second target weights corresponding to different departments.

5. The intelligent triage system based on the medical big model according to claim 1 is characterized in that: The target medical model includes: The second medical information generating unit is used to determine the Pareto optimal solution corresponding to each target patient by using a multi-objective optimization algorithm, and generate the target triage information corresponding to each target patient according to the Pareto optimal solution.

6. The intelligent triage system based on a medical big model according to any one of claims 1 to 5, characterized in that: The information acquisition module further includes: An information adjustment unit is used to determine whether the confidence of the target triage information is less than a preset confidence threshold. If the confidence of the target triage information is less than the preset confidence threshold, the target triage information is adjusted to obtain corresponding adjusted triage information.

7. The intelligent triage system based on the medical big model according to claim 6 is characterized in that: The intelligent triage system further includes: The model training module is used to iteratively train the target medical model using the target training set and the elastic weight consolidation algorithm to update the model parameters of the target medical model if the confidence of the target triage information is less than the preset confidence threshold.

8. An intelligent triage method based on a large medical model, characterized in that: Applicable to computer devices, including: Obtain multimodal medical record data corresponding to the target patient; Analyzing each of the multimodal medical record data according to the data type corresponding to each of the multimodal medical record data, and fusing each of the multimodal medical record data according to the corresponding analysis results to obtain fused medical record data; Utilize pre-defined application programming interfaces to obtain real-time influenza data, regional climate information, and target operational data, including department load, medical staff attendance, and drug inventory status. Inputting the fused medical record data, the real-time influenza data, the regional climate information, and the target operation data into a target medical big model, so that the target medical big model generates target triage information corresponding to each target patient based on the fused medical record data, the real-time influenza data, the regional climate information, and the target operation data; wherein the target medical information includes waiting time, medical department, and medical advice; The target triage information corresponding to each target patient output by the target medical model is obtained.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor is used to execute the computer program to implement the intelligent triage method based on the medical big model as claimed in claim 8.

10. A computer-readable storage medium, characterized in that Used to store a computer program, which, when executed by a processor, implements the intelligent triage method based on a large medical model as described in claim 8.

Images