Cross-border telemedicine system and method based on 6G and AI

Abstract

The invention relates to the technical field of wisdom medical treatment, and discloses a cross-border remote medical system and method based on 6G and AI, the method is based on a 6G network, the cross-border medical system integrates satellite communication into 6G mobile communication, global seamless coverage is achieved, and in addition, the cross-border remote medical system can be applied to the field of intelligent medical treatment. Under the linkage support of a global satellite positioning system, a telecommunication satellite system, an earth image satellite system and a 6G ground network, an efficient and accurate cross-border medical resource scheduling scheme is generated in a self-adaptive mode, rapid scheduling is achieved, the medical resource utilization rate is increased, and a solution is provided for cross-border telemedicine.

Description

Technical Field

[0001] The present invention relates to the field of smart medical technology, and more specifically, to a cross-border telemedicine system and method based on 6G and AI. Background Art

[0002] In the field of cross-border telemedicine, the combination of 6G and AI has the potential to bring revolutionary changes. Through the 6G network, medical data can be quickly transmitted and processed to provide support for remote diagnosis and treatment. AI technology can generate personalized treatment plans based on the patient's specific condition and historical data to improve treatment effectiveness. Intelligent monitoring equipment can collect patients' physiological data in real time, and the AI ​​system can analyze this data and issue warnings in time to reduce the occurrence of emergencies. Virtual doctor assistants can provide patients with 24-hour consulting services, reduce the workload of doctors, and improve the efficiency of medical services. 6G's global coverage makes cross-border medical cooperation possible. AI can help medical experts in different regions share knowledge and experience and jointly solve complex medical problems.

[0003] With the development of 6G communication technology and artificial intelligence generated content (AIGC) technology, cross-border telemedicine services have become an important development direction of the medical industry. Cross-border telemedicine can break through geographical restrictions, realize the sharing of high-quality medical resources, and provide timely and efficient diagnosis and treatment services for patients in remote areas. However, the existing medical resource allocation and scheduling methods still have some shortcomings and cannot meet the needs of cross-border telemedicine.

[0004] The patent application with publication number CN112509684A discloses a medical resource allocation method and system based on big data. This method obtains information such as hospital location, age distribution of permanent residents, historical medical records, etc., combines the medical resource allocation model and the distribution of the nature of work of the permanent population, generates a medical allocation plan, and adjusts the medical configuration of the hospital. However, this method is mainly aimed at the resource allocation of a single hospital, and lacks a cross-institutional and cross-regional medical resource collaborative scheduling mechanism. In addition, this method does not make full use of emerging information technologies such as 6G, and its real-time performance and flexibility need to be improved.

[0005] The patent application with publication number CN118782216A discloses a smart allocation method for hospital wards based on the Internet of Things. This method uses IoT sensing devices to collect ward data, classify patient conditions, evaluate ward resource utilization, and intelligently allocate wards based on patient recovery conditions. Although this method can improve the efficiency of resource utilization within the hospital, it is still limited to a single hospital and does not involve the problem of medical resource scheduling in cross-border telemedicine scenarios. At the same time, this method lacks comprehensive consideration of key medical factors such as doctors and drugs, making it difficult to provide patients with a full range of diagnosis and treatment plans.

[0006] To sum up, the existing technology in the field of cross-border telemedicine lacks a cross-institutional and cross-regional coordinated scheduling mechanism for medical resources, does not fully utilize emerging technologies such as 6G and AIGC, lacks real-time and flexibility, and lacks comprehensive consideration of medical elements such as doctors, equipment, and drugs, making it difficult to generate the optimal combination of medical resources. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, the present invention provides a cross-border telemedicine system and method based on 6G and AI, which uses 6G and AI technology to adaptively generate efficient and accurate cross-border medical resource scheduling solutions according to the patient's dynamic medical needs and the real-time status of medical resources.

[0008] To achieve the above object, the present invention provides the following technical solutions: Cross-border telemedicine methods based on 6G and AI include: Obtain the first state data of medical resources through the 6G terminals deployed in the medical institutions; obtain the comprehensive scheduling vector of medical resources according to the first state data of medical resources; and divide the medical resources into N scheduling levels according to the comprehensive scheduling vector of medical resources, where N is a positive integer; Receive the medical needs of patients, extract key information and perform semantic analysis on the medical needs to obtain a patient portrait reflecting the severity of the patient's condition; based on the patient portrait, select the doctors, medical equipment and drugs most relevant to the patient's condition from different scheduling levels to build a medical resource combination; Using AIGC technology, the optimal medical resource scheduling plan is generated based on the combination of medical resources.

[0009] Furthermore, the first status data of the medical resources includes information on the professional expertise of doctors, information on the work saturation status of doctors, hospital location information, real-time status information of medical equipment, and drug information; The obtaining of the comprehensive scheduling vector of medical resources according to the first state data of the medical resources includes: Analyze the first status data of medical resources and extract key scheduling indicators; the key scheduling indicators include doctor's professional expertise, work saturation, hospital location, equipment status and drug supply; Quantitatively score each key scheduling indicator to obtain a quantitative score of the corresponding key scheduling indicator; the quantitative score of the key scheduling indicator includes a doctor's professional expertise score, a work saturation score, a hospital location score, an equipment status score, and a drug supply score; Each key scheduling indicator and the quantitative score of the key scheduling indicator are combined into a comprehensive scheduling vector of medical resources.

[0010] Furthermore, the quantitative scoring of each key scheduling indicator to obtain the quantitative scoring of the corresponding key scheduling indicator includes: According to the professional expertise information of the doctor, the professional expertise of the doctor is quantitatively scored to obtain the professional expertise score of the doctor; the professional expertise information of the doctor includes the doctor's academic qualifications, years of practice and professional qualifications; According to the doctor's work saturation status information, the doctor's work saturation degree is evaluated to obtain a work saturation score; the doctor's work saturation status information includes the current number of patients and the doctor's maximum daily average number of patients; Obtain patient location information, and calculate the hospital location score based on the patient location information and hospital location information; Calculating a device status score based on real-time status information of the medical device; the real-time status information of the medical device includes the working status, load capacity and failure rate of the medical device; The drug information includes the inventory, demand and procurement cycle of the drugs; the inventory score and demand score are calculated based on the inventory and demand of the drugs; the cycle score is calculated based on the procurement cycle; the inventory score, demand score and cycle score are weighted and summed to obtain the drug supply score.

[0011] Furthermore, dividing the medical resources into N scheduling levels includes: The comprehensive scheduling vector of medical resources is normalized, and the components of each dimension are mapped to the interval [0,1]. The similarity between any two comprehensive scheduling vectors of medical resources is calculated to construct a medical resource similarity matrix. All medical resource similarity matrices are clustered to divide all medical resources into N scheduling levels with different priorities.

[0012] Furthermore, the medical needs are unstructured text information, including chief complaint, medical history, examination and medication information; The patient portrait reflecting the severity of the patient's condition includes: Extracting key information tuples according to the unstructured text information; Mapping the key information tuple to a preset medical knowledge graph and electronic medical record corpus, calculating the semantic similarity between the key information tuple and known diseases, and obtaining a disease-centered patient feature vector; According to the disease-centric patient feature vector, key information elements of the patient are extracted, the key information elements are input into a pre-trained disease severity determination model, and the disease severity level is output; the key information elements of the patient and the disease severity level are organized into a structured patient portrait.

[0013] Furthermore, the construction of the medical resource combination includes: Extract keywords from patient portraits as search terms; use the search terms to perform semantic search and relevance sorting among medical resources at different scheduling levels to obtain a sequence of candidate medical resources; Filtering an available subset of medical resources from a sequence of candidate medical resources according to the comprehensive scheduling vector of medical resources; Arrange and combine the doctors, medical equipment and drugs in the medical resource subset, calculate the comprehensive utility scores of different combinations; take the combination with the highest comprehensive utility score as the medical resource combination.

[0014] Furthermore, the step of screening out an available subset of medical resources from the candidate medical resource sequence includes: If the doctor's work saturation score is lower than the preset saturation threshold, the doctor will be included in the subset of available medical resources; If the device status score is higher than a preset device status threshold, the medical device is included in the available medical resource subset; If the drug supply score is higher than the preset drug supply threshold, the drug is included in the subset of available medical resources.

[0015] A cross-border telemedicine system based on 6G and AI, which is used to implement the above-mentioned cross-border telemedicine method based on 6G and AI, comprises: Vector construction module: obtain the first status data of medical resources through 6G terminals deployed in medical institutions; obtain the comprehensive scheduling vector of medical resources based on the first status data of medical resources; Scheduling level division module: According to the comprehensive scheduling vector of medical resources, medical resources are divided into N scheduling levels, where N is a positive integer; Patient portrait construction module: receiving the patient's medical needs, extracting key information and performing semantic analysis on the medical needs, and obtaining a patient portrait reflecting the severity of the patient's condition; Resource combination module: based on the patient portrait, select the doctors, medical equipment and drugs most relevant to the patient's condition from different scheduling levels to build a medical resource combination; Scheduling plan generation module: Using AIGC technology, the optimal medical resource scheduling plan is generated based on the combination of medical resources.

[0016] An electronic device comprises a memory, a central processing unit, and a computer program stored in the memory and executable on the central processing unit, wherein the central processing unit implements the above-mentioned cross-border telemedicine method based on 6G and AI when executing the computer program.

[0017] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the above-mentioned cross-border telemedicine method based on 6G and AI.

[0018] Compared with the prior art, the present invention has the following beneficial effects: Through the 6G high-speed, low-latency, and large-bandwidth network transmission characteristics, the present invention can quickly obtain real-time dynamic data of medical institutions, including information such as doctors' professional expertise, work saturation, hospital location, equipment status, and drug supply, providing a timely, complete, and accurate data basis for subsequent medical resource scheduling. The multi-dimensional medical resource status data is used to construct a comprehensive scheduling vector, and the medical resources are hierarchically clustered based on vector similarity to form scheduling levels with different priorities; this approach can fully consider multiple influencing factors, more comprehensively and objectively evaluate the availability and adaptability of medical resources, and make subsequent resource matching and scheduling more accurate and efficient. AIGC technology is used to perform semantic understanding and feature extraction on the patient's unstructured medical demand text, construct a structured patient portrait, and quantify the key information elements and severity of the patient's condition; based on medical big data such as knowledge graphs, AIGC can intelligently and quickly analyze the patient's condition and reduce the doctor's diagnostic burden. The hierarchically clustered medical resources are intelligently matched with the quantified patient portrait to screen out the doctor, equipment, and drug combination most relevant to the patient's condition. This patient-driven precise resource matching method can meet patients' medical needs to the greatest extent, reduce resource waste and patient waiting time, and significantly improve the quality and efficiency of medical services. Using AIGC technology to automatically generate the optimal medical resource scheduling plan can realize the integration and coordination of cross-regional medical resources, break the limitations of geography and medical conditions, promote the sharing and sinking of high-quality medical resources, and alleviate the problem of uneven distribution of medical resources. On the whole, the present invention innovatively applies 6G and AIGC to the field of cross-border telemedicine, which better solves the problems of difficulty and low efficiency in scheduling traditional medical resources. Patients can enjoy a more convenient and high-quality medical experience, and medical institutions can achieve a significant increase in resource utilization and operational efficiency, providing new ideas and new momentum for the development of smart medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 This is a principle flow chart of the cross-border telemedicine method based on 6G and AI in the present invention; Figure 2A flow chart of a method for obtaining a comprehensive scheduling vector of medical resources according to first state data of medical resources in a cross-border telemedicine method based on 6G and AI of the present invention; Figure 3 A flow chart of a method for quantitatively scoring each key scheduling indicator in the cross-border telemedicine method based on 6G and AI of the present invention to obtain a quantitative score of the corresponding key scheduling indicator; Figure 4 A flowchart of a method for dividing medical resources into N scheduling levels in the cross-border telemedicine method based on 6G and AI of the present invention; Figure 5 A flowchart of a method for obtaining a patient portrait reflecting the severity of a patient's condition in a cross-border telemedicine method based on 6G and AI of the present invention; Figure 6 A flow chart of a method for screening out a subset of available medical resources from a sequence of candidate medical resources in the cross-border telemedicine method based on 6G and AI of the present invention; Figure 7 A flow chart of a method for generating an optimal medical resource scheduling solution based on medical resource combination in the cross-border telemedicine method based on 6G and AI of the present invention; Figure 8 This is a functional module diagram of the cross-border telemedicine system based on 6G and AI in the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0022] Example 1

[0023] See also Figure 1 As shown, this embodiment provides a cross-border telemedicine method based on 6G and AI, including: Step S1000, obtaining first state data of medical resources through a 6G terminal deployed in a medical institution; obtaining a comprehensive scheduling vector of medical resources according to the first state data of the medical resources; and dividing the medical resources into N scheduling levels according to the comprehensive scheduling vector of medical resources, where N is a positive integer; Furthermore, step S1000 includes: Step S1100, obtaining first status data of medical resources through a 6G terminal deployed in a medical institution; the first status data of medical resources includes professional expertise information of doctors, work saturation status information of doctors, hospital location information, real-time status information of medical equipment, and drug information; The doctor's professional expertise information includes the doctor's education background, years of practice and professional qualifications; The doctor's work saturation status information includes the current number of patients and the doctor's maximum daily number of patients; The real-time status information of the medical device includes the working status, load capacity and failure rate of the medical device; The drug information includes the inventory, demand and procurement cycle of the drug.

[0024] Specifically, a high-speed 6G private network is deployed in each hospital, connecting smart terminals and building an information perception system within the hospital. A 6G private network refers to a local area network built in a specific area (such as a hospital) using 6G technology. Compared with the public 6G network, a 6G private network can provide faster, safer and more reliable network services. By utilizing the large bandwidth and low latency characteristics of the 6G network, massive amounts of medical data scattered everywhere can be collected in real time and efficiently, and quickly transmitted back to the intelligent scheduling platform. The large bandwidth characteristics of the 6G network enable it to support data transmission rates of billions of bits per second, and the low latency characteristics enable end-to-end latency to be controlled at the millisecond level. These characteristics ensure the real-time collection and transmission of medical data.

[0025] The 6G network of this application can be deployed in a large number of base stations, combined with spatial multiplexing technology, through the two key technologies of massive antenna array (Massive MIMO) and beamforming. At the transmitting end, the high-speed data stream is divided into multiple lower-speed sub-data streams, and different sub-data streams are transmitted on different transmitting antennas in the same frequency band. Because the spatial sub-channels between the antenna arrays of the transmitting and receiving ends are sufficiently different, the receiver can distinguish these parallel sub-data streams without paying additional frequency or time resources, and can increase channel capacity and improve spectrum utilization without occupying additional bandwidth and consuming additional transmission power. At the same time, combined with beamforming technology, the beam is managed and controlled through an algorithm to make it more focused on signal coverage of 6G terminals. Furthermore, based on the integrated air-space coverage of the 6G network, the cross-border telemedicine method based on 6G and AI in this application will greatly improve the quality and efficiency of medical services, and achieve multiple effects such as real-time and efficient data transmission, personalized treatment plans, intelligent monitoring and early warning, virtual doctor assistants, cross-border medical cooperation, optimization of medical resource allocation, improvement of patient satisfaction and support for public health decision-making.

[0026] Doctors' professional expertise can be directly obtained through electronic files. Electronic files are systems that use computers and network technology to digitally collect, store, manage and utilize various types of doctor's file information. By accessing the doctor's electronic file database, you can quickly obtain information such as the doctor's education, years of practice, and professional qualifications.

[0027] The doctor's work saturation reflects the doctor's busyness. By obtaining the number of patients currently being treated by the doctor in real time and combining it with the doctor's maximum daily average capacity, the doctor's work saturation can be evaluated. The doctor's maximum daily average capacity can be determined based on factors such as the grade of the hospital where the doctor works, the nature of the department, and the scope of diagnosis and treatment. When the work saturation is too high, the overload protection and task transfer mechanism should be activated, such as suspending the issuance of the doctor's number and diverting the patients to be treated to other doctors to avoid medical quality and safety risks.

[0028] Intelligent matching of hospital location information and patient location information is crucial to improving the convenience of medical treatment and shortening the time of treatment. By embedding the location interface in the medical information system of each hospital, the dispatch system can accurately obtain the hospital's latitude and longitude coordinates, traffic conditions and other location information. At the same time, the patient's location information can be collected through mobile phone positioning systems, smart wearable devices, etc. By calculating the distance between the patient and the hospital and combining road accessibility, the convenience of medical treatment can be evaluated.

[0029] The real-time status information of medical equipment directly affects whether the hospital can carry out relevant diagnosis and treatment. By installing networked sensors on large equipment such as CT, MRI, and high-end color ultrasound, key parameters such as their working status, load conditions, and fault warnings can be monitored in real time. Once a device failure occurs, engineers are automatically reminded to troubleshoot the fault in the shortest time possible, minimizing the interruption time of diagnosis. At the same time, big data analysis of equipment usage can be performed to rationally allocate equipment resources to avoid idle resources and overload operation.

[0030] Intelligent management and dynamic scheduling of drug information are important links in optimizing the allocation of medical resources. The traditional manual drug management model often has many problems such as information lag, poor circulation, and backlog of funds, and the Internet of Things technology can effectively solve these problems. By attaching radio frequency identification tags to drug packaging and networking in various drug warehouses and pharmacies in hospitals with wireless access points (AccessPoint, AP), real-time inventory of drug inventory and flow can be achieved, and the data can be uploaded to the cloud. Drug suppliers, medical insurance institutions, drug supervision departments, etc. can also achieve information interconnection and interoperability with hospitals through blockchain technology, thereby shortening the time for drug procurement, allocation, and settlement, and reducing financial and labor costs. In the event of a public health emergency such as an infectious disease outbreak, key drugs can also be quickly mobilized.

[0031] In data processing, algorithms such as Kalman filtering are used to suppress noise interference and improve data reliability. Heterogeneous data fields are normalized according to the state data packet format to facilitate subsequent data fusion calculations.

[0032] Through step S1100, the information islands within the medical institution are connected, and the scattered medical resources are incorporated into a unified perception network. 6G communication eliminates the bottleneck of data collection and transmission, making it possible to collect massive heterogeneous medical data in real time. Through data cleaning and feature extraction, the original multimodal data is processed into structured and computable state feature quantities. These high-quality data will lay the foundation for subsequent intelligent scheduling. The real-time collection of medical resource status data is the nerve endings of the entire smart medical system. It can timely and accurately perceive the resource distribution of the medical system, and provide data guarantee for the intelligent scheduling system to realize accurate configuration and rapid response of medical resources.

[0033] Step S1200, obtaining a comprehensive scheduling vector of medical resources according to the first state data of the medical resources; Furthermore, if Figure 2 As shown, step S1200 includes: Step S1210, parsing the first status data of the medical resources to extract key scheduling indicators; the key scheduling indicators include doctor's professional expertise, work saturation, hospital location, equipment status and drug supply; Step S1220, quantifying and scoring each key scheduling indicator to obtain a quantitative score of the corresponding key scheduling indicator; the quantitative score of the key scheduling indicator includes a doctor's professional expertise score, a work saturation score, a hospital location score, an equipment status score, and a drug supply score; Furthermore, if Figure 3 As shown, step S1220 includes: Step S1221, quantifying and scoring the doctor's professional expertise based on the doctor's professional expertise information to obtain the doctor's professional expertise score; For example, a doctor's professional expertise is quantitatively scored based on information such as his or her education level, years of practice, and professional qualifications.

[0034] A score of 0-100 is given based on a 100-point system, taking various factors into consideration.

[0035] Educational background weights: bachelor degree or below (0.2), master degree (0.3), doctorate or above (0.5); Weight of years of practice: less than 5 years (0.2), 5-10 years (0.3), more than 10 years (0.5); Professional qualification weights: general physician (0.2), attending physician (0.3), associate chief physician (0.4), chief physician (0.5); Final score = academic score + years of practice score + professional qualification score.

[0036] A doctor's professional ability is comprehensively determined by multiple factors such as academic qualifications, years of practice, and professional qualifications. In this step, a weighted scoring method is adopted to assign weights to each influencing factor. By summing up the scores of each item, a quantitative evaluation score of the doctor's professional expertise is obtained. The quantified professional expertise index can be directly used for mathematical modeling and computational analysis, facilitating the automatic execution of the matching scheduling algorithm by the computer. Quantifying the doctor's professional expertise into computable numerical scores lays a foundation for subsequent intelligent matching of diagnosis and treatment tasks.

[0037] Step S1222: Evaluate the doctor's work saturation degree based on the doctor's work saturation status information to obtain the work saturation score. Exemplarily, assume that the doctor's work saturation is divided into 5 levels: very idle, normal, relatively busy, very busy, and overloaded, corresponding to 0 - 4 points respectively.

[0038] Saturation D = current number of patients being treated / maximum daily patient reception capacity of the doctor; Among them, the maximum daily patient reception capacity of the doctor is an empirical value determined by the hospital level and department characteristics.

[0039] If D ≤ 0.2, it is very idle and gets 0 points; If 0.2 < D ≤ 0.5, it is normal and gets 1 point; If 0.5 < D ≤ 0.8, it is relatively busy and gets 2 points; If 0.8 < D ≤ 1, it is very busy and gets 3 points; If D > 1, it is overloaded and gets 4 points.

[0040] For example, Dr. Zhang San, an orthopedic doctor in a certain tertiary - level hospital, has a maximum daily patient reception capacity of 50 person - times, and currently has 35 patients being treated. Then D = 35 / 50 = 0.7. This doctor is in a relatively busy state and the work saturation score is 2 points.

[0041] Based on the doctor's current number of patients being treated and their maximum daily reception capacity, a work saturation model is constructed. The saturation and the grading show a monotonic correspondence relationship. By embedding a dynamic perception mechanism, the change of the doctor's work load is monitored in real - time. When the saturation threshold is exceeded, an early warning is issued in a timely manner, and overload protection is activated to divert tasks to relatively idle doctors. Considering the doctor's static expertise characteristics and dynamic load status together can achieve the dynamic balance and full utilization of doctor resources. By quantitatively evaluating the doctor's work saturation and dynamically perceiving the doctor's busy - idle state, doctor resources can be reasonably allocated during task assignment to avoid doctors working over - load.

[0042] Step S1223: Obtain the patient location information, and calculate the hospital location score based on the patient location information and the hospital location information. Exemplarily, the hospital location score P takes into account factors such as the distance between the patient and the hospital and transportation convenience, and is assumed to be divided into 4 levels.

[0043] P = a × distance score + b × convenience score; The distance score is divided into 4 levels: within 20 kilometers, 20-50 kilometers, 50-100 kilometers, and more than 100 kilometers, corresponding to 4-1 points respectively; the convenience score takes into account whether there are high-speed rail, airplanes and other means of transportation between the two places, corresponding to 4-1 points; a and b are weight factors, and the sum of the two is 1.

[0044] For example, if patient A is in Beijing and hospital B is in Tianjin, the two places are 120 kilometers apart and connected by high-speed rail, then: The distance score is 1 point and the convenience score is 4 points; Assuming the distance weight a is 0.6 and the convenience weight b is 0.4, then: Hospital B’s location score for Patient A is P=0.6×1+0.4×4=2.2 points.

[0045] Traditional medical resource scheduling often only considers the service capacity of the hospital, and pays less attention to the travel costs of patients. This step introduces the location scoring indicator, which can give priority to hospitals that are close to patients and have convenient transportation when matching doctors and patients, minimize the distance for patients to seek medical treatment, and increase the chances of patients receiving timely diagnosis and treatment. The weighted scoring model is used to more accurately evaluate the convenience of medical treatment by combining the distance between the two places and the convenience of transportation. Especially for critical and severe diseases, the location score is related to the patient's life safety. Introducing the location factor between patients and hospitals can improve the convenience of patients' medical treatment and shorten the treatment time for critical and severe patients.

[0046] Step S1224, calculating a device status score according to the real-time status information of the medical device; The equipment status score Q comprehensively considers the working status (S), load capacity (L) and failure rate (F) of the medical equipment, assuming a percentage system: Q = c × S × (d × L + e × (1-F)); Among them, c is the working status weight, d is the load capacity weight, and e is the failure rate weight; the sum of the three is 1, which can be determined according to the equipment type, importance, etc.; S takes values ​​between 0 and 100, indicating the current working status of the device, with 100 being the best and 0 being the worst. If the device is currently in a faulty state, S=0, and the device status score Q is also 0. L ranges from 0 to 100, indicating the load capacity of the equipment, with 100 being full load and 0 being zero load; F takes values ​​between 0 and 1, indicating the equipment failure rate, 0 means never fails, and 1 means always fails.

[0047] Based on the three indicators of equipment status, load capacity, and failure rate, a weighted evaluation model is constructed to calculate the equipment status score. The introduction of equipment status score can comprehensively evaluate whether the equipment can carry out diagnosis and treatment work normally. Poor equipment status will cause task delays or interruptions. Embedding equipment scores into the comprehensive resource scheduling model can match the best combination, improve hardware resource utilization, and reduce the queuing time for diagnosis and treatment tasks. Obtain medical equipment operating status data in real time and quantify it into equipment status scores to create conditions for the reasonable matching of hardware resources for diagnosis and treatment tasks.

[0048] Step S1225, calculate the inventory score and demand score based on the inventory and demand of the drugs; calculate the cycle score based on the procurement cycle; and perform weighted summation of the inventory score, demand score and cycle score to obtain the drug supply score.

[0049] Drug supply score Consider the inventory of medicines , Demand , Procurement cycle , the score is based on a percentage system: ; Where f is the inventory score weight, g is the demand score weight, and h is the cycle score weight; and All three scores are The score range corresponds to five levels: 1-5.

[0050] Inventory score S1: ; Demand score S2: ; Cycle score S3: ; The upper limit of the cycle score is 100 and the lower limit is 0.

[0051] For example, assume that the inventory of a certain drug is 1,000 boxes, the average monthly demand is 1,200 boxes, and the procurement cycle is 1 month: calculate : ; Therefore: Inventory Score (20 points); calculate : ; therefore: ; Calculate cycle score: ; Assume that the three weights are ,but: ; The final supply score of the drug point.

[0052] The drug supply scoring model analyzes drug inventory, demand, and procurement cycle from three perspectives, constructs a segmented scoring function, and then weights and sums the three scores. The drug supply score reflects whether drug resources are sufficient, which is crucial to ensuring the safety of patients' medication and timely drug treatment. By embedding a scoring mechanism, early warning of drug shortage risks can be provided, and supply can be allocated in a timely manner to avoid patients "stopping treatment due to medication"; at the same time, it can also optimize procurement and inventory management, reduce capital occupation and drug expiration. Dynamic evaluation of drug supply levels can help hospitals optimize drug inventory management and ensure drug supply.

[0053] Step S1230, combining each key scheduling indicator and the quantitative score of the key scheduling indicator into a comprehensive scheduling vector of medical resources.

[0054] Specifically, the purpose of step S1200 is to analyze the medical resource status data, extract key indicators that can reflect the availability and priority of medical resources, and quantify and score these indicators, and finally form a comprehensive medical resource scheduling vector. The comprehensive medical resource scheduling vector is a multi-dimensional feature vector, each dimension corresponds to the score of a key scheduling indicator, which comprehensively describes the static attributes and dynamic status of medical resources, and is an important basis for subsequent intelligent scheduling decisions.

[0055] In step S1210, the most critical factors for scheduling optimization are identified from the massive medical resource status data, including doctor's professional expertise, work saturation, hospital location, equipment status and drug supply. These scheduling indicators cover the core elements of medical resources such as people, money and materials. Among them, doctor's professional expertise reflects the doctor's professional ability level, which is directly related to whether he can meet specific medical needs. Doctor's work saturation represents the doctor's busyness. For doctors who are overloaded, the burden should be reduced in time, and idle doctor resources should be fully utilized. The hospital location indicator needs to consider the convenience of patients' medical treatment, and give priority to hospitals that are close to patients and have convenient transportation to shorten the patient's waiting time for medical treatment. The equipment status indicator focuses on whether the medical equipment can work normally. Equipment failure and excessive load will cause medical tasks to be interrupted or delayed. The drug supply indicator weighs factors such as inventory, demand, and procurement cycle to prevent drug hoarding and shortage. By extracting key indicators, the noise and redundancy in the status data are removed, the semantic characteristics of the data are highlighted, and subsequent quantitative analysis is convenient.

[0056] Step S1220 further quantifies and scores each key scheduling indicator. A normalized scoring system is used to map each indicator to a range of 0-100 points or 1-5 levels, unify the dimensions, and lay the foundation for multi-indicator evaluation. Steps S1221-S1225 give the scoring rules for each key indicator respectively. Based on the percentage system, a reasonable range of indicator scores is set. Within the indicator items, different levels are further subdivided. For example, doctors' professional expertise is scored according to academic qualifications, years of service, and professional titles, and drug supply is evaluated from three perspectives: inventory, demand, and cycle. The scoring system fully considers the multiple influencing factors of the indicators, and tries to be as comprehensive, accurate, and detailed as possible.

[0057] At the same time, for indicators that are difficult to quantify directly, a practical and feasible grading method is proposed, which gives qualitative concepts quantitative scores and enhances the computability of indicators. For example, the location of a hospital is divided into two dimensions: distance and convenience. The distance is divided into four levels, and the convenience is also divided into four levels. Then linear weighting is used to obtain the location score. For another example, the work saturation of doctors is scored according to five levels: overload, very busy, relatively busy, normal, and very idle. The saturation and level show a monotonically increasing relationship. The grading of indicators avoids the ambiguity of conceptual semantics and refines the evaluation criteria.

[0058] After obtaining the quantitative scores of each key indicator, step S1230 arranges them in order of the indicators to form a 5-dimensional vector, namely the comprehensive scheduling vector of medical resources. This vector covers many factors of medical resources, from manpower, material resources to service radius, and comprehensively describes the static attributes and dynamic operation status of medical resources. Each dimension has a clear quantitative score, which is convenient for computer programs to perform numerical analysis and strategy optimization.

[0059] The construction of the comprehensive scheduling vector fully absorbs the expert experience and formalizes and structures it with the help of mathematical models, providing a solid mathematical foundation for the subsequent multi-objective optimization solution. From qualitative to quantitative, from experience to model, from single to comprehensive, the generation process of the scheduling vector embodies the perfect unity of medical and engineering integration, expert intelligence and machine intelligence.

[0060] For example, suppose a medical task requires a cardiologist, preferably a tertiary hospital, within 500 kilometers of the patient, with a heart stent and an ECG monitor, and anticoagulants may be used during the operation. Through state data collection, three cardiologists A, B, and C are found in the candidate hospitals.

[0061] After quantitative scoring of indicators, we get: Doctor A: Professional expertise 85 points, work saturation 2 (busy), position score 3.4, equipment status 88 points, and required drug supply 55 points; Doctor B: Professional expertise 92 points, work saturation 4 (overloaded), position score 4.2, equipment status 76 points, and required drug supply 82 points; Doctor C: Professional expertise 78 points, work saturation 1 (normal), position score 2.6, equipment status 94 points, and required drug supply 61 points.

[0062] The corresponding comprehensive scheduling vectors are: Va=[85,2,3.4,88,55]; Va is the comprehensive scheduling vector of medical resources for doctor A; Vb=[92,4,4.2,76,82]; Vb is the comprehensive scheduling vector of medical resources for doctor B; Vc=[78,1,2.6,94,61]; Vc is the comprehensive scheduling vector of medical resources for doctor C.

[0063] It can be seen from this that although Doctor B has the highest professional expertise, his workload is already overloaded; although Doctor B's hospital is closest to the patients, the local drug supply is limited; the work intensity of Doctors A and C is relatively reasonable, but Doctor C's hospital is too far from the patients.

[0064] Through the construction of the comprehensive scheduling vector, the above information is quantitatively represented, and the differences and trade-offs between different indicators are reflected in the numerical value of the vector. The scheduling algorithm can calculate the weighted distance, angle cosine, etc. of the vector and select the scheduling solution with the best comprehensive indicator.

[0065] The comprehensive scheduling vector of medical resources cleverly balances the priorities between different medical resource elements and avoids the limitations that may be caused by single indicator decision-making. It realizes the digital representation of medical resources at a higher level of abstraction, describes the complex internal relationship of medical resources in mathematical language, and provides a common data interface for subsequent multi-objective optimization and intelligent scheduling.

[0066] Through the comprehensive scheduling vector and its quantification process, this solution introduces mathematical logic and computational modeling on the basis of traditional empirical scheduling, realizes the precise allocation of medical resources, and improves the pertinence and effectiveness of supply and demand matching. Compared with single and static resource management, this solution can dynamically perceive the real-time status of medical resources, and respond quickly and balance the multiple demands of both doctors and patients, reducing blind allocation and resource waste, and maximizing the efficiency of medical resource utilization.

[0067] Step S1300, dividing the medical resources into N scheduling levels according to the comprehensive scheduling vector of medical resources, where N is a positive integer.

[0068] Furthermore, if Figure 4 As shown, step S1300 includes: Step S1310, normalizing the comprehensive scheduling vector of medical resources, and mapping each dimensional component to the interval [0,1]; Step S1320, calculating the similarity between any two medical resource comprehensive scheduling vectors and constructing a medical resource similarity matrix; Step S1330: clustering all medical resource similarity matrices, and dividing all medical resources into N scheduling levels with different priorities.

[0069] Specifically, since the dimensions and numerical ranges of different indicators in the comprehensive scheduling vector of medical resources vary greatly, in order to eliminate this incomparability, the original data needs to be normalized. Commonly used normalization methods include linear normalization, logarithmic normalization, inverse tangent normalization, etc. Normalization eliminates the dimensional influence of different indicators and achieves "homogenization" of indicators. The normalized scheduling vector reflects the relative advantages and disadvantages of medical resources in various indicators, rather than the comparison of absolute values.

[0070] The similarity of the comprehensive scheduling vector of medical resources can be measured by Euclidean distance, cosine similarity, etc. The similarity matrix is ​​introduced to quantitatively characterize the similarity between different medical resources. The higher the similarity, the closer the comprehensive service capabilities of the two medical resources are, and they have a certain degree of substitutability in scheduling optimization. Resources with lower similarity have larger differences in service capabilities and should be divided into different priority levels in hierarchical scheduling. The similarity matrix is ​​the basis for cluster analysis.

[0071] For example, assuming that there are 4 comprehensive scheduling vectors of medical resources, the normalized comprehensive scheduling vectors are: V1'=(0.2, 0.9, 0.5, 0.1, 0.7); V1' is the normalized comprehensive scheduling vector of resource 1; V2'=(0.3, 0.8, 0.4, 0.2, 0.6); V2' is the normalized comprehensive scheduling vector of resource 2; V3'=(0.6, 0.4, 0.8, 0.7, 0.3); V3' is the normalized comprehensive scheduling vector of resource 3; V4'=(0.5, 0.6, 0.1, 0.8, 0.4); V4' is the normalized comprehensive scheduling vector of resource 4; Then the similarity matrix S' of the medical resource comprehensive scheduling vector is (keep 2 decimal places): ; From the matrix, we can see that the similarity between resource 1 and resource 2 is 0.22, which is the smallest among the non-diagonal elements, indicating that they are the most similar; the similarity between resource 3 and resource 4 is 0.67, which is the second smallest non-diagonal element besides 0.22, so their similarity is second only to resource 1 and resource 2. This similarity matrix provides a distance measurement basis for subsequent clustering.

[0072] The K-means clustering algorithm is used to cluster the medical resource similarity matrix. The K-means clustering algorithm achieves clustering by minimizing the sum of squares of distances between objects within each cluster and the cluster center. The K-means clustering algorithm is used to automatically group medical resources. The comprehensive service capabilities of resources within each group are relatively close, while the differences between groups are large. The algorithm minimizes the sum of squares of distances within the group through iterative optimization to achieve the goal of high similarity within the group and low coupling between groups. The final clustering result corresponds to N scheduling levels from high to low. It is worth noting that the selection of the N value should be moderate, which should not only ensure the differentiation of priorities, but also avoid the levels being too discrete and reducing the flexibility of scheduling.

[0073] Assume N=3, the clustering result is (G1,G2,...,G N ), G N The clustering result corresponding to the scheduling level of priority N is shown, and the four medical resources are finally divided into three priority levels: G1={resource 1, resource 2}, G2={resource 4}, G3={resource 3}; Among them, G1 has the highest priority, G2 is the second, and G3 is the lowest. This classification result takes into account the comprehensive performance of medical resources in multi-dimensional attributes, and can avoid the one-sidedness caused by prioritizing based on only one indicator.

[0074] In short, step S1300 uses data mining technology to fully integrate the multi-dimensional attribute information of medical resources, characterize the differences in resource endowments, and automatically generate a scheduling priority sequence, laying a solid mathematical foundation for subsequent resource optimization configuration. At the same time, this process is adaptive, and the classification results can be dynamically updated according to changes in actual conditions. Hierarchical scheduling is conducive to alleviating the contradiction between supply and demand of medical resources, allowing high-quality resources to serve critically ill patients first, and fully mobilizing idle resources to participate in routine diagnosis and treatment, thereby maximizing the service potential of medical resources.

[0075] Step S2000, receiving the medical needs of the patient, extracting key information and performing semantic analysis on the medical needs, and obtaining a patient portrait reflecting the severity of the patient's condition; based on the patient portrait, selecting the doctor, medical equipment and medicine most relevant to the patient's condition from different scheduling levels, and constructing a medical resource combination; Furthermore, step S2000 includes: Step S2100, receiving the medical needs of the patient, extracting key information and performing semantic analysis on the medical needs, and obtaining a patient portrait reflecting the severity of the patient's condition; Furthermore, if Figure 5 As shown, step S2100 includes: Step S2110, receiving the patient's medical needs, where the medical needs are unstructured text information, including chief complaint, medical history, examination and medication information; Specifically, patients fill out relevant forms on the mobile app, describing their physical signs, past medical history, allergy history, family history, and medical examinations and medications they have taken in detail. They can also upload multimedia information such as pictures and videos related to their condition. These unstructured data require further semantic analysis and feature extraction, while also protecting patient privacy and desensitizing sensitive information.

[0076] For example, suppose patient A submits the following medical needs: He has felt chest tightness and shortness of breath in the past two weeks, and the symptoms worsen after a little activity. He has a history of hypertension and poor blood pressure control. He had a coronary angiography in the hospital last year and was diagnosed with coronary heart disease. He is currently taking oral drugs such as Locinda and Bayer, but the symptoms have not been significantly relieved.

[0077] Step S2120, extracting key information tuples according to the unstructured text information; Specifically, the unstructured text information is processed by word segmentation, part-of-speech tagging, named entity recognition, etc., to extract key information tuples such as patient complaints, symptoms and signs, medical history and medication, and examination results; For example, for the medical needs text of patient A, the following key information tuples can be extracted: 〈Main complaint: chest tightness and shortness of breath〉; 〈Symptoms and signs, symptoms aggravated by activity〉; 〈Patient history, history of hypertension〉; 〈Living conditions, poor blood pressure control〉; 〈Examination results, coronary angiography, coronary heart disease〉; 〈Current medications: Locinda, Bayer Aspirin〉; 〈The effect of medication, no significant relief of symptoms〉; Among them, the original text is processed by machine learning models such as word segmentation algorithms based on medical vocabulary and conditional random fields, and key words and phrases related to the patient's condition are identified and classified according to their parts of speech and contextual semantics to form structured information tuples. These information tuples cover the key elements of the patient's condition, but are still discrete and non-quantitative representations that require further analysis and refinement.

[0078] Step S2130, mapping the key information tuple to a preset medical knowledge graph and electronic medical record corpus, calculating the semantic similarity between the key information tuple and known diseases, making a preliminary judgment on the disease suffered by the patient, and obtaining a patient feature vector centered on the disease; Specifically, in the medical knowledge graph, each disease corresponds to a node, and the symptoms, signs, examinations, medications, etc. related to the disease also constitute nodes, and there are semantically related edges between the disease nodes and these nodes. The knowledge graph represents the multidimensional characteristics of the disease in a structured way. At the same time, a large number of electronic medical records are desensitized to form a corpus, which records the manifestations of various diseases in the real world and reflects the diversity of the disease. The patient key information tuple obtained in step S2120 is semantically matched with the knowledge graph and the corpus respectively. Taking the patient's symptom data as an example, in the knowledge graph, the semantic similarity between the patient's symptoms and the candidate disease nodes is calculated using medical ontology and semantic embedding technology, and the N1 most relevant diseases are selected; in the corpus, text matching algorithms such as BM25, TF-IDF, etc. are used to retrieve the medical records with the highest similarity to the patient's symptoms, and the distribution of each disease is statistically analyzed. Based on the similarity results of knowledge graph matching and corpus matching, the candidate diseases are weighted and sorted, and the most likely disease is preliminarily identified.

[0079] Repeat the above process until all components in the patient's key information tuple complete semantic matching. On this basis, further calculate the overall similarity between the patient's key information tuple and the candidate disease, and select the final disease discrimination result. Connect the disease with the patient's symptoms, signs, examinations, medication and other information to form a subgraph as the patient's disease-centered feature vector. This vector depicts the manifestation of the disease from different aspects, contains rich semantic information, and is the basis for subsequent severity judgment.

[0080] For example, patient Zhang San has symptoms of fever and cough, and a routine blood test shows an elevated white blood cell count. First, match in the knowledge graph, calculate the semantic similarity of {fever, cough, elevated white blood cell count} with each disease node, and find that it is highly correlated with diseases such as pneumonia, upper respiratory tract infection, and sepsis; then match in the corpus, and retrieve a large number of pneumonia medical records containing the above symptoms, while there are fewer similar medical records for other diseases. Combining the semantic similarity of the two, it is preliminarily determined that Zhang San is most likely to have pneumonia. Further match Zhang San's age, gender, medical history and other information with the typical characteristics of pneumonia, and the overall similarity is very high, so pneumonia is diagnosed. Finally, a feature vector is formed with pneumonia as the central node and fever, cough, elevated white blood cells, etc. as peripheral nodes, and the vector also contains other information about Zhang San.

[0081] The semantic matching and similarity calculation in step S2130 make full use of the medical semantic knowledge contained in the knowledge graph and corpus. Compared with simple retrieval matching, semantic technology can reveal the deep connection between medical concepts and more accurately identify the patient's disease. Matching the patient's overall information with the disease overcomes the limitations of a single dimension, comprehensively considers the various manifestations of the disease, and improves the accuracy of the identification. The patient feature vector integrates the disease ontology knowledge and the patient's individual characteristics. It is a highly semantic abstraction of the patient's condition and can meet the needs of subsequent intelligent diagnosis and treatment.

[0082] Through semantic matching and similarity calculation, the disease knowledge most relevant to the patient is "sifted" from the massive medical knowledge and medical record data to form a semantically rich feature vector. This semantically driven patient representation simulates the doctor's thinking process and realizes the mapping from the patient's appearance to the essence of the disease. Compared with the traditional "symptom-disease" correspondence method, the semantic method can handle complex and diverse medical data, has stronger generalization and extrapolation capabilities, and is an important step for smart healthcare to move towards in-depth understanding.

[0083] Step S2140, extracting the key information elements of the patient based on the disease-centric patient feature vector, inputting the key information elements into a pre-trained disease severity determination model, and outputting the disease severity level; organizing the key information elements of the patient and the disease severity level into a structured patient portrait.

[0084] Specifically, based on the disease-centered patient feature vector, key information elements such as patient age, gender, course of disease, time of initial diagnosis, main symptoms, important signs, and laboratory test results are extracted; although the disease-centered patient feature vector covers the main aspects of disease manifestations, its form is still mainly in the form of graphs and texts, which is not convenient for quantitative judgment of severity. Therefore, it is necessary to further refine it into structured key information elements. For quantitative indicators such as age, course of disease, blood pressure, heart rate, body temperature, etc., the numerical values ​​are directly extracted; for qualitative indicators such as symptoms, the relevant sentences are normalized into a standard vocabulary, such as "moderate moist rales can be heard in the right lower lung" is converted to "pulmonary rales". Remove ambiguity in expression and unify the caliber of indicators.

[0085] The disease severity determination model adopts a deep learning architecture, and common ones include Bidirectional Encoder Representations from Transformers (BERT), eXtra Long Network (XLNet), Enhanced Representation throughkNowledge Integration (ERNIE), etc. The model is pre-trained with massive text data on massive medical knowledge graphs, electronic medical record corpora, and clinical guidelines to acquire deep semantic knowledge of medical language; fine-tuning is done on vertical medical field data to master the specialized skills of disease severity determination. Multi-channel disease representations are input, including symptoms, signs, auxiliary examinations, medication, etc., to fully understand the condition and output the severity level. The model can be trained end-to-end and automatically learn complex mappings from input to output.

[0086] Specifically, the training process of the disease severity determination model is as follows: Data preparation: Collect massive amounts of medical text data such as medical knowledge graphs, electronic medical record corpora, clinical guidelines, and perform structured processing on them to extract structured fields such as symptoms, signs, auxiliary examinations, medication, and severity.

[0087] Example: Electronic medical record 1: Symptoms: cough, sputum, wheezing, fever (38.5°C); Physical signs: moist rales in the right lower lung, heart rate 110 beats / min, respiration 30 times / min; Auxiliary examination: chest CT showed right lower lung infection; Medication: Ceftriaxone, Jianfei mixture; Severity: Moderate; Clinical Guideline 1: Disease: Community-acquired pneumonia; Mild: mild symptoms, body temperature <38°C, no dyspnea, heart rate <100 beats / min; Moderate: obvious symptoms, body temperature ≥38°C, dyspnea, heart rate ≥100 beats / min; Severe: severe symptoms, arterial oxygen saturation <90%, and signs of septic shock; Annotated data: Manual annotation is performed on some structured corpora, that is, clinicians determine the severity of the disease based on symptoms, signs and other information, which serves as the gold standard for the training set and validation set. The annotation form is a (text, severity) tuple.

[0088] Example: {Cough for 3 days, coughing up yellow sputum, body temperature 37.8°C, white blood cell count (WBC) 12*10 9 / L, C-reactive protein (CRP) 30 mg / L, common cold}, mild; {Continuous high fever for 5 days, body temperature 39.5°C, cough with purulent sputum, chest pain, dyspnea, chest CT showed large shadows in the middle and lower lobes of the right lung, severe pneumonia}, severe; Text representation: Preprocess the medical text, such as word segmentation, stop word removal, lowercase conversion, etc., and map the words into vectors. Common word vector models include the word vector model (Word2Vec) and the global vector model (GloVe). Represent each text as a word vector sequence and input it into the model.

[0089] Pre-training: Use pre-trained language models such as BERT to perform unsupervised pre-training on large-scale medical text corpora to acquire deep knowledge of medical language, such as lexical, grammatical, and semantics. Pre-training tasks include: Masked language model: randomly mask some words and predict these masked words; Next sentence prediction: predict whether two sentences are adjacent to each other; Example: Input: persistent high fever for [masked] days, body temperature [masked]°C, cough with purulent sputum. Prediction: [5], [39.5].

[0090] Input: Continuous high fever for 5 days, temperature 39.5°C. Coughing up purulent sputum, chest pain, dyspnea. Prognosis: Yes.

[0091] Fine-tuning: Based on pre-training, supervised fine-tuning of the model is performed using labeled data to enable it to learn the task of determining the severity of the disease. The specific approach is to concatenate symptoms, signs, auxiliary examinations, etc. into a text sequence, input it into the BERT encoder, and then connect it to the severity classifier to perform end-to-end multi-classification. The loss function uses cross entropy, and the optimization algorithm uses Adam.

[0092] Example: Input: Continuous high fever for 5 days, body temperature 39.5°C, cough with purulent sputum, chest pain, dyspnea, chest CT showed large shadows in the middle and lower lobes of the right lung; Output: Heavy.

[0093] Model application: For new patient case texts, extract key information elements such as symptoms, signs, and auxiliary examinations, input them into the trained disease severity determination model, and output the severity level of the patient's disease.

[0094] Example: Input: Male patient, 65 years old, had fever for 3 days, body temperature 39°C, accompanied by severe cough, rusty sputum, chest pain, dyspnea, and extensive moist rales in both lungs. Chest CT showed diffuse patchy shadows and interstitial changes in both lungs. Blood routine: WBC 15*10 9 / L, neutrophil ratio 0.85, CRP 150mg / L.

[0095] Extract key information elements: Age: 65 years old Gender: Male; Symptoms: fever, severe cough, coughing up rusty sputum, chest pain, and difficulty breathing; Physical signs: temperature 39°C, moist rales in both lungs; Auxiliary examination: chest CT showed diffuse patchy shadows and interstitial changes in both lungs; WBC 15*109 / L, neutrophil percentage (NE%) 0.85; CRP 150mg / L; Output: Heavy; In short, the training process of the disease severity determination model includes: preparing a large amount of medical text corpus, manually annotating a small amount of training data, preprocessing and representing medical texts, pre-training the model to acquire medical language knowledge, and fine-tuning the model to acquire disease severity determination skills. The input is structured information such as the patient's symptoms, signs, and auxiliary examinations, and the output is the disease severity level. By automatically and intelligently determining the severity of the disease, doctors can be assisted in quickly evaluating and triaging the patient's condition, rationally allocating medical resources, and optimizing the treatment process. Human-machine collaboration is expected to improve the treatment effect of serious diseases.

[0096] Patient portrait example, organizes the information of patient Zhang San and the severity level of the disease pneumonia "severe" into a structured patient portrait record <Name: Zhang San, Age: 45, Gender: Male, Disease: Pneumonia, Severity Level: Severe>.

[0097] Through step S2140, structured key information elements are further extracted from the disease-centric patient feature vector, sent to the pre-trained model to determine the severity of the disease, and qualitative results are output to form a structured and labeled patient portrait. The portrait describes the patient's basic condition and disease severity in a structured manner, providing an explainable and operational basis for the subsequent formulation of precise diagnosis and treatment plans. The pre-trained model automatically performs feature extraction and severity mapping through deep learning, which can be generalized to a variety of diseases, saving a lot of manual analysis costs.

[0098] Patient portraits are an important part of the smart medical system. The traditional medical service model is "hospital-centered" and lacks consideration of individual patient characteristics, resulting in mismatch of medical resources. Patient portraits are patient-centered, describing patients from multiple angles such as demography, symptomatology, physiology, and pathology, allowing the medical system to "understand patients", which is the basis for achieving precise and personalized diagnosis and treatment. This step integrates a variety of AI technologies such as medical knowledge graphs, electronic medical records, and deep learning to construct a comprehensive, accurate, and intelligent patient portrait, which promotes the upgrade of medical services from extensive to refined.

[0099] Step S2200, selecting doctors, medical equipment and drugs most relevant to the patient's condition from different scheduling levels according to the patient portrait, and constructing a medical resource combination; Further, step S2200 includes: Step S2210, extracting keywords of the patient portrait as search terms; using the search terms to perform semantic search and relevance sorting in medical resources at different scheduling levels to obtain a candidate medical resource sequence; First, extract keywords from the patient portrait as search terms. The patient portrait is a comprehensive description of all patient information, including demographic characteristics, physiological characteristics, medical history, chief complaints, examination results, etc. Natural language processing technology is used to segment and tag the patient portrait, and identify keywords such as symptoms, signs, and disease names. For example, for the case of "male patient, 45 years old, acute myocardial infarction, heart rate 110 beats / min, accompanied by chest tightness and sweating", keywords such as "male", "45 years old", "acute myocardial infarction", "heart rate 110", "chest tightness", and "sweating" can be extracted.

[0100] Next, the keywords are used as search terms to perform semantic search in the medical resource knowledge base. Medical resources are divided according to the scheduling level, and search indexes are established in the knowledge base for resources of different levels. Semantic search not only considers the literal matching of keywords, but also uses knowledge enhancement methods such as medical ontology and word vectors to explore the semantic associations between words. For example, although "myocardial infarction" and "coronary heart disease" are different in words, they are similar in meaning and should be matched during search. In addition, word sense disambiguation is also required. For example, the word "rheumatism" refers to different diseases in "rheumatic heart disease" and "rheumatoid arthritis", and it needs to be distinguished according to the context during search.

[0101] Semantic retrieval obtains a series of candidate medical resources related to the patient's condition, including doctors, equipment, drugs, etc. In order to further screen out the best resources, the candidate resources need to be sorted by relevance. Relevance sorting uses a variety of similarity algorithms to calculate the similarity between patients and candidate resources from the perspectives of disease symptoms, pathology, treatment, etc. Take two indicators as an example: the number of symptom overlaps and the disease pathology correlation. The more the disease symptoms overlap between the patient and the candidate resource, or the higher the similarity of the pathogenesis of the two diseases, the stronger the correlation. According to the weighted sum of various indicators, the relevance value of the candidate resource can be obtained. Sorting from high to low according to the relevance value, that is, the candidate medical resource sequence is obtained. The medical resources ranked higher in the sequence have a higher match with the patient and are more likely to provide high-quality and efficient diagnosis and treatment services.

[0102] Step S2210 uses semantic retrieval and relevance sorting to quickly and accurately screen out a subset of resources that are highly relevant to the patient's condition from a large amount of heterogeneous medical resources, thereby narrowing the search scope of subsequent resource combination optimization and improving combination efficiency and quality. Semantic retrieval breaks through the limitations of keyword matching, and mines the implicit association between doctors and patients at the semantic level of vocabulary and concepts, making the retrieval results more comprehensive and accurate. Multi-index weighted relevance evaluation reflects the diversity of patient needs and resource endowments, such as the severity of the disease, doctor expertise, geographical distance, etc., which will affect the comprehensive relevance of doctor-patient matching. Qualitative relevance requirements are characterized by quantitative similarity scores, providing a computable objective function for resource combination optimization.

[0103] Step S2220, screening out a subset of available medical resources from a sequence of candidate medical resources according to the comprehensive scheduling vector of medical resources; Furthermore, if Figure 6 As shown, step S2220 includes: Step S2221, if the doctor's work saturation score is lower than a preset saturation threshold, the doctor is included in the available medical resource subset; Step S2222, if the device status score is higher than a preset device status threshold, the medical device is included in the available medical resource subset; Step S2223: If the drug supply score is higher than a preset drug supply threshold, the drug is included in the available medical resource subset.

[0104] Specifically, step S2221 introduces the concept of doctor work saturation threshold. The saturation threshold refers to the upper limit of the doctor's workload. If it exceeds this threshold, the doctor's work intensity will be unbearable, he will be easily fatigued, and the service quality will also decline. Therefore, only doctors whose work saturation is lower than the threshold will be regarded as available medical resources. The determination of the threshold needs to balance resource utilization and medical risks. Too high will lead to doctor overload, and too low will cause resource waste. Generally, the correspondence between saturation score and classification can be formulated according to factors such as hospital level and department nature. For example, the threshold of the intensive care unit of a tertiary hospital can be set to 1, that is, doctors are not allowed to be "very busy"; while the threshold of the general outpatient clinic of a grassroots hospital can be relaxed to 3, allowing "overload". When the doctor's work saturation score is lower than the threshold of the position, it means that he still has surplus service capacity. The system will include him in the available doctor resource pool. When receiving new diagnosis and treatment tasks, he will be matched and scheduled from this resource pool first. By setting the saturation threshold reasonably, we can make full use of doctor resources and avoid excessive fatigue of doctors. While ensuring the efficiency of medical services, we can also take into account medical quality and safety.

[0105] The device status threshold in step S2222 refers to the minimum status score required for the normal and stable operation of the medical device. The device status score comprehensively considers factors such as whether the device can be turned on, the current load level, and the historical failure rate. The setting of the device status threshold should balance the intensity of use and reliability. If the threshold is set too low, the risk of equipment failure will increase, and if it is too high, the equipment will be idle. The thresholds of different types of equipment are not consistent. The status requirements of large equipment such as nuclear magnetic resonance and CT are more stringent, generally reaching more than 80 points; while the threshold of small equipment such as stethoscopes can be relaxed to 60 points. When the status score of a device is higher than the threshold of the device type to which it belongs, it indicates that its performance is stable and its operation is reliable, and it can be used to undertake new inspection or treatment tasks. At this time, the system will automatically add the device to the list of available medical devices, and when scheduling tasks, it will be selected from the available devices first. Medical equipment is an important material basis of modern medicine. Reasonable assessment of its status and dynamic inclusion in the available resource pool can improve equipment utilization and shorten the waiting time for patients' inspections and treatments. At the same time, the status threshold can timely detect potential faults and avoid equipment working with illness.

[0106] Step S2223 proposes the concept of drug supply threshold. The supply threshold reflects the minimum requirement for drug supply and demand balance. Only drugs with supply scores higher than the threshold have sufficient inventory to meet clinical drug demand. The level of the threshold is directly related to whether patients can obtain the required drugs in time. Too low may cause drug shortages and delay treatment; too high will lead to capital occupation and drug backlog. The reasonable setting of the threshold needs to take into account the characteristics of the drug dosage form, indications, and procurement cycle. The threshold for commonly used drugs such as antibiotics can be set at 80 points, and the threshold for small-variety drugs such as rare disease drugs can be relaxed to 60 points. For some critical life-saving drugs, a "one-vote veto" rule can also be set. As long as any score of inventory or procurement cycle is 0, even if the comprehensive score meets the standard, it cannot be included in the list of available drugs. When the supply score of a drug is higher than the threshold of the drug category to which it belongs, it means that its inventory is relatively sufficient and can meet clinical use needs in the short term. At this time, it will enter the hospital's available drug catalog. When prescribing for patients, doctors can directly select available drugs from the catalog without worrying about the risk of "out of stock". The supply threshold reflects the patient's accessibility to medication and is one of the important indicators of pharmaceutical management. Timely inventory and update of the available status of drugs can improve the safety, continuity and economy of patients' medication.

[0107] Step S2220 sets the availability thresholds of resource elements such as doctor saturation, equipment status, and drug supply, and dynamically screens the available resource subset accordingly. Its significance lies in establishing an automatic resource optimization mechanism, identifying high-quality resource combinations with strong service capabilities and low risk of receiving patients from the massive combination of medical resources at any time, incorporating them into the available resource pool, and giving priority matching recommendations during the intelligent scheduling process. This dynamic resource management mode based on threshold judgment can maximize the efficiency of medical resources while avoiding resource overload, and ensure the quality and safety of medical services. For example, when the work saturation of a surgeon exceeds the standard and a key device in the operating room is in a faulty state, the system can automatically block the doctor and the device, and instead match the patient with a doctor with lower saturation and an operating room with intact equipment. For another example, when the inventory of a commonly used drug in the hospital is in urgent need and the supply score is close to the threshold, the system will issue an early warning and recommend that the pharmacy department initiate an emergency procurement plan to ensure that the patient's medication needs are not affected. It can be foreseen that with the continuous maturity and improvement of the dynamic adjustment mechanism of thresholds, the ability to flexibly allocate medical resources will be greatly enhanced, reducing resource mismatch and waste, and providing patients with better-quality and more efficient smart medical services.

[0108] Step S2230, arrange and combine the doctors, medical equipment and drugs in the medical resource subset, calculate the comprehensive utility scores of different combinations; and take the combination with the highest comprehensive utility score as the medical resource combination.

[0109] Specifically, first, for the selected subset of medical resources, all combinations of doctors, equipment, and drug resources are listed by permutation and combination. The combination process follows the basic principle of "applicable and optimal", that is, the resource combination model is matched with the patient's basic diagnosis and treatment process. Taking the diagnosis and treatment model of "acute myocardial infarction-coronary intervention" as an example, it is necessary to combine resource elements such as "emergency department, cardiology department-cardiac catheterization room-thrombolytic drugs, stents". Different combination modes represent a series of alternative diagnosis and treatment plans for patients. With the continuous deepening of medical subdivisions, the combination mode and scale are becoming more and more complex, and a comprehensive enumeration of combination plans places high demands on computing power.

[0110] Secondly, a comprehensive utility evaluation index is constructed to score and rank each alternative combination scheme. Utility is an economic concept, which specifically refers to the effect and value that the medical resource combination scheme brings to patients. The dimensions of comprehensive utility include diagnostic accuracy, efficacy, complication rate, patient comfort, cost, etc. Taking efficacy and complications as an example, if a combination scheme can quickly control the disease, improve prognosis, and has a low incidence of adverse reactions, its efficacy utility score should be high. In order to balance multiple utility dimensions, it is also necessary to assign weights to indicators in each dimension, such as efficacy weight 0.4, complication weight 0.3, comfort weight 0.2, and cost weight 0.1, and then weighted sum to obtain the comprehensive utility score. Utility weights reflect the relative importance of different indicators to resource combination decisions and can be adjusted according to the personalized needs of patients. Finally, the one with the highest comprehensive utility score is selected from many alternative combinations as the optimal medical resource combination scheme. The optimal scheme achieves global maximization in the multi-indicator utility score, balances the medical effects obtained by patients and the diagnosis and treatment resources used, and represents the optimal solution for the diagnosis and treatment combination under the current state.

[0111] Step S2230 integrates and optimizes multiple medical resources with the goal of maximizing utility. In the traditional medical model, there is a serious asymmetry of information between doctors and patients, and resource allocation lacks a quantitative basis. Introducing comprehensive utility into resource combination optimization can objectively measure the input-output ratio of different diagnosis and treatment combinations, so that patients can obtain medical resources fairly and justly. At the same time, the multi-indicator utility evaluation that combines weights and scores uses a clear and concise comprehensive score to reflect the cost-effectiveness ranking of the plan. The personalized needs of patients can also be achieved by adjusting the utility weights. For example, patients who value efficacy more can increase the efficacy weight, and patients who care more about costs can increase the cost weight. Resource combination optimization embeds the rational choices of both supply and demand sides into the algorithm model, allowing patients and doctors to reach a consensus on diagnosis and treatment.

[0112] Step S3000, using AIGC technology to generate an optimal medical resource scheduling plan based on medical resource combination.

[0113] Furthermore, if Figure 7 As shown, step S3000 includes: Step S3100, using AIGC technology to generate an initial scheduling solution set based on medical resource combination; Specifically, the combination of medical resources is used as the input of the AIGC model, and large language models such as Transformer are used to output a series of possible initial scheduling solutions through autoregressive generation. These initial solutions cover different resource combinations and priority rankings, exploring the solution space as widely as possible. The generation process makes full use of the powerful semantic understanding and generation capabilities of the pre-trained language model, making the solution rich in content, logically coherent, professional and reliable.

[0114] Step S3200, performing secondary optimization screening on the initial scheduling solution set according to the comprehensive scheduling vector and scheduling level of medical resources to generate an optimized scheduling solution set; The matching degree between the initial scheduling plan and the medical resource vector is calculated. Specifically, the scheduling plan can also be represented as a multidimensional vector, and its dimensions correspond to key scheduling indicators such as doctor's professional expertise, work saturation, hospital location, equipment status, and drug supply. Then, algorithms such as cosine similarity are used to calculate the similarity between the scheduling plan vector and the comprehensive scheduling vector of medical resources to obtain the matching degree between the two. The higher the matching degree, the more consistent the scheduling plan is with the current medical resource status, and the higher the resource utilization rate is, and it should be given priority.

[0115] Select the solution that best matches the current status of medical resources. At the same time, refer to the scheduling level of resources, give priority to high-priority resources, and further screen the scheduling solution. This process can be regarded as a constrained search on the initial solution set. The objective function of the search is to maximize resource utilization and optimize priority matching. The search space is limited to the resource constraint boundary represented by the vector. The optimized and screened solution is guaranteed in terms of feasibility and scheduling quality.

[0116] Step S3300, using a reinforcement learning algorithm, iteratively optimizes the optimization scheduling solution set in a simulation environment to obtain an optimal medical resource scheduling solution.

[0117] Specifically, the set of optimized scheduling schemes is used as the initial state set of reinforcement learning, and each scheme corresponds to an initial state. The entire execution process of the scheme is simulated in a virtual simulation environment. By setting the agent, state space, action space and reward function, the agent learns the optimal scheduling strategy. Among them, the agent can be regarded as a scheduling decision maker, the state space contains the dynamic changes of medical resources, the action space covers various scheduling control operations, and the reward function comprehensively considers multiple optimization goals such as patient satisfaction, medical resource utilization, and task completion efficiency. The advantages of reinforcement learning over supervised learning are: 1) it does not require a large amount of labeled data, and learns autonomously through interaction with the environment; 2) it can discover the optimal strategy, not limited to empirical decision-making; 3) it is also applicable to continuous state space and action space. Reinforcement learning has many applications in the field of resource scheduling, such as power grid scheduling, vehicle path planning, etc., which can provide useful reference for this scenario.

[0118] After multiple rounds of iterations, the reinforcement learning algorithm finally converges to the optimal scheduling strategy. Applying this strategy to the optimized scheduling solution set can obtain the dynamically optimal medical resource scheduling solution. This solution responds to changes in medical needs in real time, accurately matches and dynamically optimizes resources, and achieves multi-objective balanced optimization of medical resource utilization efficiency, task completion quality, and operating cost savings while meeting patients' medical needs.

[0119] Step S3000 innovatively applies AIGC technology to medical resource scheduling, giving full play to the technical advantages of artificial intelligence in sequence generation, combinatorial optimization, and strategy learning. First, large language models such as Transformer are used to generate initial plans. The pre-training of massive knowledge makes the generated content professional and diverse, simulating the decision-making ideas of human experts to a certain extent. On this basis, secondary optimization screening is carried out in combination with task constraints to further improve the adaptability and feasibility of the plan. Finally, reinforcement learning is introduced to iteratively optimize the plan, overcome the limitations of the rule system, and realize the adaptive dynamic optimization of the scheduling strategy. The strategy evaluation and improvement are realized through the simulation environment, avoiding the risk of blind attempts in real scenarios, and providing a strong guarantee for the implementation of the scheduling plan.

[0120] The beneficial effects of step S3000 include: 1) improving the comprehensive utilization efficiency of medical resources and alleviating the problem of shortage and idleness of high-quality resources; 2) enhancing the accuracy of supply and demand matching, reducing patients' ineffective waiting and running around, and improving the medical experience; 3) optimizing the fairness of scheduling, allocating resources based on medical needs and treatment time, and eliminating human intervention and regional differences; 4) ensuring the quality of task completion, comprehensively considering factors such as doctor's expertise, equipment performance, and drug supply, and providing a high-quality and efficient combination of diagnosis and treatment solutions for difficult and complex diseases; 5) realizing intelligent and adaptive scheduling, flexibly responding to emergency incidents, and improving emergency response capabilities.

[0121] For example, patient A suddenly develops acute appendicitis, and the positioning system shows three nearby hospitals, namely A, B, and C. The patient demand information and the resource combination vector of the three hospitals are input into the AIGC system, and 10 initial scheduling plans are generated by autoregression, mainly involving the permutation and combination of elements such as the consultation time, the consultation department, the designated expert, the operation time, and the type of medicine. Then, according to the real-time updated resource comprehensive vector and the scheduling level, the initial plan is optimized, and 5 plans are retained. Then, in the emergency scheduling simulation environment, the state space, action space, and reward function of reinforcement learning are set. The state space contains dynamic information such as hospital road conditions, number of beds, and operating room occupancy. The action space includes scheduling measures such as selecting a hospital for treatment, designating experts, arranging surgery time, and configuring drug packages. The reward function is quantified from dimensions such as medical convenience, surgical timeliness, and drug delivery speed. After 500 iterations of training, the optimal scheduling strategy in this case is obtained: Patient A goes to the nearest hospital B, diagnosed by chief physician Zhang San, and arranges emergency surgery at 18:30 on the same day, using the hospital's regular drug package. The entire consultation process was compressed into 5 hours, saving nearly 3 hours compared with the traditional experience-based scheduling. This scheduling solution has achieved breakthroughs in terms of intelligence, response speed, and process optimization, winning precious time for the rescue of critically ill patients.

[0122] Example 2 This embodiment provides a cross-border telemedicine method based on 6G and AI on the basis of Embodiment 1, including: Step S2230, arrange and combine the doctors, equipment, and medicines in the medical resource subset, calculate the comprehensive utility scores of different combinations; and take the combination with the highest comprehensive utility score as the medical resource combination.

[0123] The calculation of the comprehensive utility scores of different combinations includes: ; The meanings of the parameters are as follows: : The comprehensive utility score of a combination, ranging from , a higher score indicates greater utility.

[0124] E c : Efficacy score, which reflects the effect of the combination on disease control and prognosis improvement, can be obtained through meta-analysis (meta-analysis) or network meta-analysis, and the value range is .

[0125] C c : The incidence of complications, which reflects the risk of adverse reactions of the combination, can be obtained through prospective cohort studies and ranges from .

[0126] S c : The patient's quality of life score reflects the patient's physiological, psychological, and social adaptability after receiving the combination. It can be evaluated using the SF-36 scale with a value range of .

[0127] F c : Cost burden score, which reflects the proportion of the out-of-pocket expenses of the patient using this medical resource combination to his / her family income. It can be estimated through questionnaire surveys and medical bills, and the value range is .

[0128] : Doctor-patient relationship score, which reflects the patient's satisfaction with the medical staff's communication and service, can be assessed using the PDRQ-9 scale, with a range of .

[0129] T c : Treatment duration score, which reflects the length of time required for the patient to receive the medical resource combination, can be estimated based on the clinical pathway and follow-up, and the value range is .

[0130] :Efficacy-complication interaction adjustment coefficient, reflecting the sensitivity of efficacy to complication risk adjustment. Increased complication rates When the fluctuation occurs, adjust The size of controls the impact of this change on the overall utility. The lower the value, the more effective the treatment will be.

[0131] :Nonlinear weight adjustment coefficient of complications, which controls the weight of complications in the whole utility formula in a sinusoidal form, and makes a sinusoidal adjustment to the impact of complication risk, so that it has a smaller impact on the utility score when the risk is low, and becomes more influential when the complication risk increases. The value will amplify the nonlinear effect of complications and make the formula more sensitive to changes in complications.

[0132] :Quality of life-cost interaction adjustment coefficient, adjust the sensitivity of quality of life score to cost burden, high The value of 0 indicates that the quality of life score is more significantly affected by the cost burden, thus affecting the overall utility. , which can reflect the patient's sensitivity to cost expenditure, especially in the pursuit of high quality of life. This parameter can ensure that resource allocation is more inclined to economically reasonable options.

[0133] : Cost burden nonlinear weight adjustment coefficient, which controls the weight of the cost burden score in cosine form. This coefficient performs cosine adjustment on the interaction between cost burden and quality of life, ensuring that it has a smaller impact on the utility score when the cost burden is low, and a greater impact when the cost burden is high. The value can reflect the nonlinear impact of changes in cost burden on quality of life.

[0134] : Doctor-patient relationship-negative impact adjustment coefficient, which controls the weight of the doctor-patient relationship score in the negative adjustment of the utility score. The value will amplify the negative impact of poor doctor-patient relationship on the utility score, making the formula more sensitive to the quality of patient experience, thereby paying more attention to human factors in decision-making. It ensures that when the doctor-patient relationship score is low, the logarithmic term of this factor in the denominator will increase rapidly, thereby reducing the overall utility.

[0135] :The index adjustment coefficient of the doctor-patient relationship, adjusts the negative index amplification effect of the doctor-patient relationship on the utility score, and adjusts , we can more accurately control the impact of the improvement or deterioration of the doctor-patient relationship on the utility. The value makes the change in doctor-patient relationship more significant, reflecting the significant contribution of continued improvement in patient relationship improvement to the overall utility score.

[0136] :Treatment duration-negative adjustment coefficient, controlling the sensitivity of treatment duration score to the negative impact of utility score, high The value amplifies the weight of the treatment time in the arc tangent function of the denominator, so that when the treatment time increases, the impact on the utility score is more significant, which is suitable for treatment scenarios with high time sensitivity. This parameter can dynamically control the impact of treatment time on the score in the denominator, thereby reflecting the importance of timeliness in resource combination.

[0137] : The nonlinear adjustment coefficient of treatment time controls the nonlinear amplification degree of treatment time score. , we can control the inverse tangent adjustment of the treatment duration on the utility score, making the formula more flexible to the changes in duration. When it is larger, the duration of treatment has a greater impact on the utility score, ensuring a reasonable assessment of the effect of long-term treatment.

[0138] : The adjustment coefficient of each parameter reflects the interaction strength between factors. It can be fitted by structural equation model or neural network model, and both are positive real numbers.

[0139] The formula adopts nonlinear relationships and uses a variety of nonlinear function forms such as trigonometric functions, logarithmic functions, exponential functions, and inverse tangent functions to construct a complex nonlinear relationship between the six factors of efficacy, complications, quality of life, cost burden, doctor-patient relationship, and treatment duration, breaking through the limitations of traditional linear combinations. The interaction terms between parameters are as follows and It respectively reflects the regulation of complications by therapeutic efficacy and the regulation of cost burden by quality of life, captures the synergistic or antagonistic effects among factors, and makes the comprehensive utility score closer to clinical reality.

[0140] Dynamic regulation of negative effects: natural logarithm ln and arc tangent are introduced in the denominator function to dynamically regulate the negative impact of doctor-patient relationship and treatment duration on comprehensive utility. When the doctor-patient relationship deteriorates or the treatment duration is too long, ln and The value of will increase rapidly, resulting in a decrease in the comprehensive utility score, which more sensitively reflects the importance of patient experience.

[0141] All are positive values ​​to ensure that the direction of influence of each factor is consistent with clinical expectations, which is determined by technical personnel in this field based on a large number of experiments. When the scores of efficacy, quality of life, and doctor-patient relationship increase, the comprehensive utility score increases accordingly; when the scores of complications, cost burden, and treatment duration increase, the comprehensive utility score decreases accordingly.

[0142] This formula comprehensively evaluates the comprehensive utility of medical resource combinations from six dimensions: efficacy, safety, quality of life, cost, doctor-patient relationship, and timeliness. By setting various parameters reasonably, the utility preferences in different clinical scenarios can be better fitted. When the efficacy and quality of life improve, the complications and cost burden decrease, the doctor-patient relationship improves, and the treatment time is shortened, the comprehensive utility score obtained by this formula will be significantly improved. Therefore, this formula can assist medical decision-making, select the diagnosis and treatment combination with the highest comprehensive utility, and realize the precise allocation of medical resources while improving patients' sense of gain, which has significant clinical application value.

[0143] Example 3 This embodiment provides a cross-border telemedicine system based on 6G and AI on the basis of Embodiment 1. Figure 8 As shown, including: Vector construction module: obtain the first status data of medical resources through 6G terminals deployed in medical institutions; obtain the comprehensive scheduling vector of medical resources based on the first status data of medical resources; Scheduling level division module: According to the comprehensive scheduling vector of medical resources, medical resources are divided into N scheduling levels, where N is a positive integer; Patient portrait construction module: receiving the patient's medical needs, extracting key information and performing semantic analysis on the medical needs, and obtaining a patient portrait reflecting the severity of the patient's condition; Resource combination module: based on the patient portrait, select the doctors, medical equipment and drugs most relevant to the patient's condition from different scheduling levels to build a medical resource combination; Scheduling plan generation module: Using AIGC technology, the optimal medical resource scheduling plan is generated based on the combination of medical resources.

[0144] In the vector construction module, the first status data of the medical resources includes the professional expertise information of the doctor, the work saturation status information of the doctor, the location information of the hospital, the real-time status information of the medical equipment and the drug information; The doctor's professional expertise information includes the doctor's academic qualifications, years of practice and professional qualifications; the doctor's work saturation status information includes the current number of patients and the doctor's maximum daily number of patients; the real-time status information of the medical equipment includes the working status, load capacity and failure rate of the medical equipment; the drug information includes the inventory, demand and procurement cycle of the drugs.

[0145] The obtaining of the comprehensive scheduling vector of medical resources according to the first state data of the medical resources includes: Step S1210, parsing the first status data of the medical resources to extract key scheduling indicators; the key scheduling indicators include doctor's professional expertise, work saturation, hospital location, equipment status and drug supply; Step S1220, quantifying and scoring each key scheduling indicator to obtain a quantitative score of the corresponding key scheduling indicator; the quantitative score of the key scheduling indicator includes a doctor's professional expertise score, a work saturation score, a hospital location score, an equipment status score, and a drug supply score; Step S1230, combining each key scheduling indicator and the quantitative score of the key scheduling indicator into a comprehensive scheduling vector of medical resources.

[0146] The quantitative scoring of each key scheduling indicator to obtain the quantitative scoring of the corresponding key scheduling indicator includes: Step S1221, quantifying and scoring the doctor's professional expertise based on the doctor's professional expertise information to obtain the doctor's professional expertise score; Step S1222, evaluating the doctor's work saturation degree according to the doctor's work saturation status information to obtain a work saturation score; Step S1223, obtaining patient location information, and calculating a hospital location score based on the patient location information and hospital location information; Step S1224, calculating a device status score according to the real-time status information of the medical device; Step S1225, calculate the inventory score and demand score based on the inventory and demand of the drugs; calculate the cycle score based on the procurement cycle; and perform weighted summation of the inventory score, demand score and cycle score to obtain the drug supply score.

[0147] In the scheduling level division module, the medical resources are divided into N scheduling levels according to the comprehensive scheduling vector of medical resources, including: Step S1310, normalizing the comprehensive scheduling vector of medical resources, and mapping each dimensional component to the interval [0,1]; Step S1320, calculating the similarity between any two medical resource comprehensive scheduling vectors and constructing a medical resource similarity matrix; Step S1330: clustering all medical resource similarity matrices, and dividing all medical resources into N scheduling levels with different priorities.

[0148] In the patient portrait construction module, the patient portrait reflecting the severity of the patient's condition includes: Step S2110, receiving the patient's medical needs, where the medical needs are unstructured text information, including chief complaint, medical history, examination and medication information; Step S2120, extracting key information tuples according to the unstructured text information; Step S2130, mapping the key information tuple to a preset medical knowledge graph and electronic medical record corpus, calculating the semantic similarity between the key information tuple and known diseases, making a preliminary judgment on the disease suffered by the patient, and obtaining a patient feature vector centered on the disease; Step S2140, extracting the key information elements of the patient based on the disease-centric patient feature vector, inputting the key information elements into a pre-trained disease severity determination model, and outputting the disease severity level; organizing the key information elements of the patient and the disease severity level into a structured patient portrait.

[0149] In the resource combination module, according to the patient portrait, the doctor, medical equipment and medicine most relevant to the patient's condition are selected from different scheduling levels to construct a medical resource combination, including: Step S2210, extracting keywords of the patient portrait as search terms; using the search terms to perform semantic search and relevance sorting in medical resources at different scheduling levels to obtain a candidate medical resource sequence; Step S2220, screening out a subset of available medical resources from a sequence of candidate medical resources according to the comprehensive scheduling vector of medical resources; Step S2230, arrange and combine the doctors, medical equipment and drugs in the medical resource subset, calculate the comprehensive utility scores of different combinations; and take the combination with the highest comprehensive utility score as the medical resource combination.

[0150] The step of screening out an available subset of medical resources from the candidate medical resource sequence comprises: Step S2221, if the doctor's work saturation score is lower than a preset saturation threshold, the doctor is included in the available medical resource subset; Step S2222, if the device status score is higher than a preset device status threshold, the medical device is included in the available medical resource subset; Step S2223: If the drug supply score is higher than a preset drug supply threshold, the drug is included in the available medical resource subset.

[0151] In the scheduling plan generation module, the use of AIGC technology to generate the optimal medical resource scheduling plan based on the combination of medical resources includes: Step S3100, using AIGC technology to generate an initial scheduling solution set based on medical resource combination; Step S3200, performing secondary optimization screening on the initial scheduling solution set according to the comprehensive scheduling vector and scheduling level of medical resources to generate an optimized scheduling solution set; Step S3300, using a reinforcement learning algorithm, iteratively optimizes the optimization scheduling solution set in a simulation environment to obtain an optimal medical resource scheduling solution.

[0152] Example 4

[0153] This embodiment discloses an electronic device, which may include one or more processors and one or more memories. The memories store computer-readable codes, which, when executed by the one or more processors, may execute the cross-border telemedicine method based on 6G and AI as described above.

[0154] The method or system according to the implementation of the present application can also be implemented with the help of the architecture of an electronic device. The electronic device may include a bus, one or more CPUs, a read-only memory (ROM), a random access memory (RAM), a communication port connected to a network, an input / output component, a hard disk, etc. A storage device in an electronic device, such as a ROM or a hard disk, can store the cross-border telemedicine method based on 6G and AI provided in this application. The cross-border telemedicine method based on 6G and AI may, for example, include: obtaining the first state data of medical resources through a 6G terminal deployed in a medical institution; obtaining a comprehensive scheduling vector of medical resources according to the first state data of medical resources; dividing medical resources into N scheduling levels according to the comprehensive scheduling vector of medical resources, where N is a positive integer; receiving the medical needs of the patient, extracting key information and semantically analyzing the medical needs, and obtaining a patient portrait reflecting the severity of the patient's condition; according to the patient portrait, selecting the doctors, medical equipment and drugs most relevant to the patient's condition from different scheduling levels to construct a medical resource combination; using AIGC technology to generate the optimal medical resource scheduling plan based on the medical resource combination.

[0155] Furthermore, the electronic device may also include a user interface. Of course, the architecture disclosed in the present invention is only exemplary, and when implementing different devices, one or more components in the electronic device disclosed in the present invention may be omitted according to actual needs.

[0156] Example 5

[0157] This embodiment discloses a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the cross-border telemedicine method based on 6G and AI in the embodiment of the present application can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache). Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0158] In addition, according to the implementation of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application, for example: obtaining the first state data of medical resources through a 6G terminal deployed in a medical institution; obtaining a comprehensive scheduling vector of medical resources according to the first state data of medical resources; dividing medical resources into N scheduling levels according to the comprehensive scheduling vector of medical resources, where N is a positive integer; receiving the patient's medical needs, extracting key information and semantically analyzing the medical needs, and obtaining a patient portrait reflecting the severity of the patient's condition; according to the patient portrait, selecting the doctor, medical equipment and medicine most relevant to the patient's condition from different scheduling levels, and constructing a medical resource combination; using AIGC technology, generating an optimal medical resource scheduling plan based on the medical resource combination. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0159] The methods, systems, and devices of the present application may be implemented in many ways. For example, the methods, systems, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.

[0160] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0161] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A cross-border telemedicine method based on 6G and AI, characterized in that: The method comprises: Obtain the first state data of medical resources through the 6G terminals deployed in the medical institutions; obtain the comprehensive scheduling vector of medical resources according to the first state data of medical resources; and divide the medical resources into N scheduling levels according to the comprehensive scheduling vector of medical resources, where N is a positive integer; Receive the medical needs of patients, extract key information and perform semantic analysis on the medical needs to obtain a patient portrait reflecting the severity of the patient's condition; based on the patient portrait, select the doctors, medical equipment and drugs most relevant to the patient's condition from different scheduling levels to build a medical resource combination; Using AIGC technology, the optimal medical resource scheduling plan is generated based on the combination of medical resources.

2. The cross-border telemedicine method based on 6G and AI according to claim 1, characterized in that: The first status data of the medical resources includes information on the professional expertise of doctors, information on the work saturation status of doctors, hospital location information, real-time status information of medical equipment and drug information; The obtaining of the comprehensive scheduling vector of medical resources according to the first state data of the medical resources includes: Analyze the first status data of medical resources and extract key scheduling indicators; the key scheduling indicators include doctor's professional expertise, work saturation, hospital location, equipment status and drug supply; Quantitatively score each key scheduling indicator to obtain a quantitative score of the corresponding key scheduling indicator; the quantitative score of the key scheduling indicator includes a doctor's professional expertise score, a work saturation score, a hospital location score, an equipment status score, and a drug supply score; Each key scheduling indicator and the quantitative score of the key scheduling indicator are combined into a comprehensive scheduling vector of medical resources.

3. The cross-border telemedicine method based on 6G and AI according to claim 2, characterized in that: The quantitative scoring of each key scheduling indicator to obtain the quantitative scoring of the corresponding key scheduling indicator includes: According to the professional expertise information of the doctor, the professional expertise of the doctor is quantitatively scored to obtain the professional expertise score of the doctor; the professional expertise information of the doctor includes the doctor's academic qualifications, years of practice and professional qualifications; According to the doctor's work saturation status information, the doctor's work saturation degree is evaluated to obtain a work saturation score; the doctor's work saturation status information includes the current number of patients and the doctor's maximum daily average number of patients; Obtain patient location information, and calculate the hospital location score based on the patient location information and hospital location information; Calculating a device status score based on real-time status information of the medical device; the real-time status information of the medical device includes the working status, load capacity and failure rate of the medical device; The drug information includes the inventory, demand and procurement cycle of the drugs; the inventory score and demand score are calculated based on the inventory and demand of the drugs; the cycle score is calculated based on the procurement cycle; the inventory score, demand score and cycle score are weighted and summed to obtain the drug supply score.

4. The cross-border telemedicine method based on 6G and AI according to claim 1, characterized in that: The division of medical resources into N scheduling levels includes: The comprehensive scheduling vector of medical resources is normalized, and the components of each dimension are mapped to the interval [0,1]. The similarity between any two comprehensive scheduling vectors of medical resources is calculated to construct a medical resource similarity matrix. All medical resource similarity matrices are clustered to divide all medical resources into N scheduling levels with different priorities.

5. The cross-border telemedicine method based on 6G and AI according to claim 1, characterized in that: The medical needs are unstructured text information, including chief complaint, medical history, examination and medication information; The patient portrait reflecting the severity of the patient's condition includes: Extracting key information tuples according to the unstructured text information; Mapping the key information tuple to a preset medical knowledge graph and electronic medical record corpus, calculating the semantic similarity between the key information tuple and known diseases, and obtaining a disease-centered patient feature vector; According to the disease-centric patient feature vector, key information elements of the patient are extracted, the key information elements are input into a pre-trained disease severity determination model, and the disease severity level is output; the key information elements of the patient and the disease severity level are organized into a structured patient portrait.

6. The cross-border telemedicine method based on 6G and AI according to claim 3 is characterized in that: The construction of medical resource combination includes: Extract keywords from patient portraits as search terms; use the search terms to perform semantic search and relevance sorting among medical resources at different scheduling levels to obtain a sequence of candidate medical resources; Filtering an available subset of medical resources from a sequence of candidate medical resources according to the comprehensive scheduling vector of medical resources; Arrange and combine the doctors, medical equipment and drugs in the medical resource subset, calculate the comprehensive utility scores of different combinations; take the combination with the highest comprehensive utility score as the medical resource combination.

7. The cross-border telemedicine method based on 6G and AI according to claim 6, characterized in that: The step of screening out an available subset of medical resources from the candidate medical resource sequence comprises: If the doctor's work saturation score is lower than the preset saturation threshold, the doctor will be included in the subset of available medical resources; If the device status score is higher than a preset device status threshold, the medical device is included in the available medical resource subset; If the drug supply score is higher than the preset drug supply threshold, the drug is included in the subset of available medical resources.

8. A cross-border telemedicine system based on 6G and AI, which is used to implement the cross-border telemedicine method based on 6G and AI as described in any one of claims 1 to 7, characterized in that: The system comprises: Vector construction module: obtain the first status data of medical resources through 6G terminals deployed in medical institutions; obtain the comprehensive scheduling vector of medical resources based on the first status data of medical resources; Scheduling level division module: According to the comprehensive scheduling vector of medical resources, medical resources are divided into N scheduling levels, where N is a positive integer; Patient portrait construction module: receiving the patient's medical needs, extracting key information and performing semantic analysis on the medical needs, and obtaining a patient portrait reflecting the severity of the patient's condition; Resource combination module: based on the patient portrait, select the doctors, medical equipment and drugs most relevant to the patient's condition from different scheduling levels to build a medical resource combination; Scheduling plan generation module: Using AIGC technology, the optimal medical resource scheduling plan is generated based on the combination of medical resources.

9. An electronic device comprising a memory, a central processing unit, and a computer program stored in the memory and executable on the central processing unit, characterized in that: When the central processor executes the computer program, the cross-border telemedicine method based on 6G and AI described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed, implements the cross-border telemedicine method based on 6G and AI as described in any one of claims 1 to 7.

Images