Visual and intelligent auxiliary management method based on chronic disease patient GIS (Geographic Information System)

Abstract

The invention discloses a GIS (Geographic Information System) visualization and intelligent auxiliary management method based on chronic disease patients. The method comprises the following steps: (1) carrying out GIS (Geographic Information System) visualization on chronic disease patient distribution, and generating a disease aggregation degree thermodynamic diagram; (2) a hybrid screening engine which combines a GIS interaction tool with a medical health label and displays data of different granularities; (3) binding the medical health intelligent robot with a verbal skill template, presetting a trigger condition, and generating a personalized reply; and (4) after the medical health intelligent robot or model completes health consultation on the patient, automatically extracting data related to health, physical examination and biochemical indexes, updating the data to a health file GIS visual platform of a corresponding user, reading the latest data update, and displaying dynamic monitoring of the chronic disease patient in the jurisdiction in real time. The beneficial effects of the invention are that the system supports the intelligent management service of multi-round dialogues and precise medical health for specific crowds in a target area, and achieves data continuity, precise service and efficient management.

Description

Technical Field

[0001] The present invention relates to the technical field related to artificial intelligence and medical health services, and in particular to a GIS-based visualization and intelligent auxiliary management method for chronic disease patients. Background Art

[0002] The existing community health services for chronic diseases have the following problems: 1. Data dispersion: Health records, diagnosis and treatment records, and other data of chronic disease patients or residents are not interoperable across medical institutions. This results in community doctors lacking continuous patient health data support, making it difficult to carry out comprehensive risk prevention and management.

[0003] 2. Resource limitations: Community doctors manage a large number of patients. Usually, each doctor has to manage the health records of hundreds or even thousands of residents. Manual follow-up visits and medical record updates are inefficient, service coverage is insufficient, the effective update rate of electronic medical records in the jurisdiction is low, and chronic disease management updates are not urgent.

[0004] 3. Inefficient chronic disease management: Relying on manual reminders, follow-up visits, and medical record updates, the system is inefficient and difficult to scale. Bringing offline health services to the community requires a significant amount of pre-notification and preparation. Patients cannot be automatically categorized by symptom and severity, and elderly people living alone cannot receive preventive care. Summary of the Invention

[0005] The present invention aims to overcome the above-mentioned deficiencies in the prior art and provides a GIS-based visualization and intelligent auxiliary management method for chronic disease patients that achieves data continuity, accurate services and efficient management.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: A method for GIS-based visualization and intelligent auxiliary management of chronic disease patients, specifically including the following steps: (1) GIS visualization of the distribution of chronic disease patients: extracting address data from electronic health case data, projecting the disease onto a GIS map based on the longitude and latitude of the community, and generating a heat map of disease aggregation; (2) Hybrid screening engine: Combine GIS interactive tools with medical and health tags, and use a privacy computing engine to ensure data security. Through public data authorization operations, source data is stored in the secure space of the National Health Commission, and different granularity data are displayed on the terminal according to the authority of the medical institution. (3) Task Configuration Center: The medical and health intelligent robot is bound to speech templates, pre-sets trigger conditions, generates personalized responses, and supports multi-round dialogue and intent recognition; (4) Macro-management analysis: Community health centers obtain regular updates of the health and medical commission data within their jurisdiction through public data authorization, or after the community health center obtains user authorization, the user regularly fills in the health status and physical examination data through the designated APP or mini-program. It also supports the health visit of medical health intelligent robots or the import of data from intelligent health consultation models. That is, after the medical health intelligent robot or model completes the health consultation for the patient, it automatically extracts data related to health, physical examination, and biochemical indicators and updates them to the corresponding user's health file. The GIS visualization platform reads the latest data updates and displays the dynamic monitoring of chronic disease patients in the jurisdiction in real time.

[0007] This invention provides a method for providing health services to target patients using a combination of GIS interactive tools and medical health tag screening, combined with a large language model-based medical health intelligent robot. This method allows medical institutions and doctors to circle patients in a specific area using a map. By combining patient characteristics, electronic medical records, medical records, and other private data, the method uses voice simulation to implement automated return visits, medication reminders, health follow-up, medical record updates, and examination invitations. This method supports multiple rounds of dialogue and intelligent management of precision medical health services for specific populations in the target area, achieving data continuity, precise services, and efficient management.

[0008] Preferably, in step (1), the following steps are performed: desensitizing the information of chronic disease patients, retaining the characteristics of symptoms and severity, extracting the address data from the electronic health case data, taking the value of the community without including the house number, projecting the disease onto the GIS map according to the longitude and latitude of the community, and when missing values ​​are encountered, authorizing and registering the updated address through the patient, displaying different colors according to the severity of the disease, and generating a heat map of the concentration of the disease.

[0009] Preferably, in step (2), specifically: select patients within the target range through the GIS interactive tool, support the standard five-level administrative boundaries of province, city, district, county, and street, and also support customized boundary fences, select one or more communities or isochronous fences, select different granularity data according to different scenarios, and cross-screen according to the patient's medical health tags within the target fence.

[0010] As a preferred option, for medical and health labels, the specific operations are as follows: (a) In the electronic medical record, the user's address is a location tag, and patients within the target service area are circled on the GIS map. From the perspective of health prevention and control, the density of patients with different diseases in different areas can be seen; (b) setting disease labels based on common chronic disease types; (c) Based on national medical and health standards or customized conditions for special populations, high, medium-high, medium and low risk levels are established for the above-mentioned different diseases, and multi-dimensional management and services are carried out based on regional location, disease symptoms and severity.

[0011] As a preferred option for GIS interaction, GIS interactive tools refer to a series of interactive controls, operating components or functional modules embedded in the GIS map interface, which enable users to deploy in the geographic information visualization interface and support users to achieve spatial positioning and attribute screening of target populations through multi-dimensional interactive operations. Multi-dimensional interaction includes administrative area selection, custom fence drawing, multi-level screening, and real-time changes in map layers based on selections, which is equivalent to screening by offline service scope and qualified users within the scope. The specific operations are as follows: (21) The user opens the GIS map interface; (22) Open the custom fence in the toolbar and draw a geographic range by hand; (23) Select all patients whose addresses in the electronic health records are within the target geographic fence according to the geographic scope, and automatically extract patients who are both within the target service geographic scope and meet the target disease conditions by screening medical and health tags; (24) At this time, the GIS map displays the thermal distribution of patients who meet the conditions in real time, and the data statistics of patients who meet the conditions within the geographic fence are drawn; (25) Set the following tasks for the selected patient population.

[0012] Preferably, in step (4), for the medical health intelligent robot: the community health doctor authorized by the user checks and updates the health information of the corresponding user, selects the target area through the GIS visualization module according to the doctor's jurisdiction area and generates an intelligent voice access task, realizes automatic screening outbound call tasks by area, disease, severity, and special care, adjusts the speech strategy according to the real-time status of the call, combines TTS simulation and slot reading technology to make the communication process more humane, and assists the doctor in updating the health information of the contracted patient. After the call ends, the medical health intelligent robot will read the key information slot of the interactive voice content, automatically structure the inquiry result data, store it in the health file data and update the health file of the corresponding patient.

[0013] As a preferred option, the medical health intelligent robot uses multimodal fusion technology of voice signal analysis + text semantic understanding + dialogue context modeling during the interaction process to achieve real-time perception of the patient's call status and drive dynamic adjustment of the conversation strategy; the specific technical features are as follows: (41) Speech emotion recognition model training: Based on the Transformer framework training, speech is converted into text tokens, temporal information is injected through positional encoding, and global features are extracted by multi-layer encoders; a private Conformer X multimodal attention chronic disease speech emotion recognition model is trained through CNN+Transformer to achieve recognition of the six basic emotions of speech and map them to the emotional scene abstraction during the interaction process of medical and health intelligent robots; (42) Real-time call status perception: During a call, voice stream data and semantic text stream data are collected synchronously, and multi-dimensional status features are extracted through the following submodules: (421) Speech signal feature extraction: Based on the acoustic feature extraction model, time domain features, frequency domain features and non-stationary features are extracted from the original speech to form a speech emotion feature vector; (422) Text semantic sentiment analysis: fine-tuning the real-time transcribed conversation text and outputting sentiment categories; (423) Conversational context modeling: Using two-way conversations, combined with the text and state features of the past 3-5 rounds of conversations, we construct the emotion vector recognized by Conformer X in the conversational context to capture the trend of patients’ emotional changes and potential needs; (43) Real-time speech generation: Personalize the selected speech templates and achieve more natural voice interaction and accurate health data collection through emotional TTS speech synthesis and context-aware slot extraction technology.

[0014] Preferably, in step (43), for emotional TTS speech synthesis, the specific steps are as follows: (i) Training a private voice model based on community doctor voices: The medical health intelligent robot combines TTS voice to simulate the doctor's voice; (ii) Physician authorization and voice cloning and personalized adaptation: During the patient's first interaction, a 5-10 second standard voice clip is collected. Voiceprint features are extracted using a speaker embedding model and combined with a general TTS model to generate a patient-specific emotional voice cloning model, making the synthesized voice closer to the patient's familiar voice style. (iii) Ensuring multi-round dialogue coherence: The dialogue state tracking module records the goals of the current dialogue turn and automatically adds connectives during TTS synthesis to ensure smooth dialogue.

[0015] Preferably, in step (43), for context-aware slot extraction, after the call ends or during the interaction, deep semantic analysis is performed on the historical voice interaction content to extract key health information and convert it into structured data; the technical implementation includes: (A) Full conversation text cleaning: De-noise and correct sentence segments of the ASR transcribed text to generate standardized conversation text; (B) Multi-granularity slot extraction: A joint entity recognition and relation extraction model based on SpanBERT is used to extract health slot information of the following dimensions from the text: basic information slot, symptom description slot, physiological indicator slot, and medication information slot; (C) The model solves the cross-round slot association problem through the contextual attention mechanism; (D) Follow-up and completion of ambiguous information: If a key slot is detected to be missing or vaguely expressed, the system will automatically trigger follow-up dialogue during the interaction and mark the slot as "unconfirmed" after the conversation ends, pushing it to the manual review module; (E) Structured data storage: The extracted slot information is converted into structured fields through standardized mapping and written into the patient's electronic health record through the HL7 FHIR interface to ensure that the data is traceable and analyzable.

[0016] The beneficial effects of the present invention are: realizing services such as automated return visits, medication reminders, health follow-up, medical record updates and examination invitations, supporting multi-round dialogues and intelligent management services for precise medical health for specific populations in target areas, and achieving data continuity, precise services and efficient management. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a system architecture diagram of the present invention; Figure 2 It is a GIS-based management map of chronic diseases by administrative region; Figure 3 It is a GIS custom condition interactive screening map; Figure 4 It is a customized intelligent outbound call task map; Figure 5 It is a GIS visualization management and analysis chart for chronic diseases. DETAILED DESCRIPTION

[0018] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0019] like Figure 1 In the embodiment described, a method for GIS-based visualization and intelligent auxiliary management of chronic disease patients specifically includes the following steps: (1) GIS visualization of the distribution of chronic disease patients: extract the address data from the electronic health case data, project it onto the GIS map according to the longitude and latitude of the community, and generate a heat map of the disease concentration; specifically: desensitize the information of chronic disease patients, retain the characteristics such as the disease and severity, extract the address data from the electronic health case data, take the value of the community without the house number, project it onto the GIS map according to the longitude and latitude of the community, and when missing values ​​are encountered, authorize and register the address to be updated with the patient, display different colors according to the severity of the disease, and generate a heat map of the concentration of different diseases.

[0020] (2) Hybrid screening engine: Combine GIS interactive tools with medical and health tags, combine with privacy computing engines to ensure data security, save source data in the security space of the Health Commission through public data authorization operation, and display different granularity data on the terminal according to the authority of the medical institution; specifically: select patients within the target range through GIS interactive tools, support standard provincial, municipal, district, county, and street-level administrative boundaries, and also support custom boundary fences, circle one or more communities or isochronous fences, circle different granularity data according to different scenarios, and cross-screen within the target fence according to medical and health tags such as patient age, symptoms, and medication records, and visualize the distribution heat of chronic diseases in GIS, intuitively presenting the distribution density of chronic disease patients in provinces, cities, districts, counties, and streets under different management authorities, and make effective adjustments to prevention and control resources according to the concentration of symptoms, such as Figure 2 shown.

[0021] Among them, for GIS interaction, GIS (Geographic Information System Interactive Tools) interactive tools refer to a series of interactive controls, operating components or functional modules embedded in the GIS map interface, which enable users to deploy in the geographic information visualization interface and support users to achieve spatial positioning and attribute screening of target populations through multi-dimensional interactive operations. Multi-dimensional interaction includes administrative area selection, custom fence drawing, multi-level screening, and real-time changes of map layers according to selections. It is equivalent to screening by offline service scope and users who meet the conditions within the scope, to prevent the conditions from being met but the offline service scope cannot be reached, or the service scope can be covered but the target service customers are not. Specific examples are as follows: (21) The user opens the GIS map interface; (22) Select the target administrative area (e.g. Haidian District, Beijing), open "Custom Fence" in the toolbar, and draw a geographical range by hand. It can be circular, rectangular, polygonal or isochronous (the maximum range under different options of walking, cycling and driving to reach the center point selected on the map at different times. The purpose is to see the effective range covered by different means of transportation at different times with the center point selected on the map, and the influence of road network, rivers and elevated roads needs to be considered). Figure 3As shown; (23) Through the above screening, all patients whose addresses in the electronic health records are within the target geographic fence are selected according to the geographical scope. At this time, check "Age 60+ + Diabetes + High Risk" in the right filter panel, that is, by filtering the medical health tags, patients who are both within the target service geographic scope and meet the target disease conditions are automatically extracted; (24) At this time, the GIS map displays the thermal distribution of patients who meet the conditions in real time, and displays the data statistics of patients who meet the conditions within the drawn geographic fence on the left; (25) Supports setting tasks for selected patient groups, such as exporting reports or executing intelligent voice outbound calls.

[0022] Among them, for medical and health labels, the specific operations are as follows: (a) In electronic medical records, the user's address is a location tag, and patients within the target service area are circled on the GIS map. From a health prevention and control perspective, the GIS map can also be used to see the density of patients with different diseases in different areas. (b) setting disease labels based on common chronic disease types, such as hypertension, diabetes, coronary heart disease, stroke, chronic obstructive pulmonary disease, malignant tumors, chronic kidney disease, osteoporosis, rheumatoid arthritis, and hyperlipidemia; (c) Based on national medical and health standards or customized criteria for specific populations, high, medium, and low risk levels are established for each of the above-mentioned conditions, enabling multi-dimensional management and services based on regional location, condition, and severity. For example, in the case of an infectious disease outbreak, the target patient concentration area is first identified, and then the distribution density of different severity levels is determined. The relationship between location and transmission is significant. This is an example of the relationship between disease, severity, and GIS location.

[0023] (3) Task configuration center: The medical health intelligent robot is bound to speech templates, such as medication reminders, follow-up questionnaires, etc., and trigger conditions are pre-set, such as the third day after surgery, to generate personalized responses, support multi-round dialogue and intention recognition, such as Figure 4 As shown; (4) Macro-management analysis: Community health centers obtain regular updates of the health commission data within their jurisdiction through public data authorization, or after the community health center obtains user authorization, users regularly fill in health status and physical examination data through designated APPs and mini-programs. It also supports data import from medical health intelligent robots for health visits or intelligent health consultation models. That is, after the medical health intelligent robot or model completes the health consultation for the patient, it automatically extracts data related to health, physical examination, and biochemical indicators and updates them to the health records of the corresponding users. Since the data of the health commission and the patients are updated irregularly, the GIS visualization platform reads the latest data updates and displays dynamic monitoring of chronic disease trends, electronic medical record update rates, community health doctor contract rates, and outbound call task execution progress in real time, such as Figure 5 shown.

[0024] Among them, for medical and health intelligent robots: community health doctors authorized by users view and update the health information of corresponding users, circle the target area through the GIS visualization module according to the doctor's jurisdiction area and generate intelligent voice access tasks, realize automatic screening outbound call tasks by area, disease, severity, and special care, adjust the speech strategy according to the real-time status of the call, combine TTS simulation and slot reading technology to make the communication process more humane, and assist doctors in updating the health information of contracted patients, return visits, care and other tasks. After ending the call, the medical and health intelligent robot will read the key information slot of the interactive voice content, automatically structure the inquiry result data, store it in the health file data and update the corresponding patient's health file.

[0025] Existing intelligent voice interaction systems typically adjust speech based solely on keyword matching or simple semantic rules, failing to perceive patient emotional fluctuations or conversation context. This results in stilted interactions and inefficient information collection. This is particularly true in the medical field, where the recognition and linking of many specialized terms, as well as the significant impact of patient emotions on communication effectiveness, are crucial. The intelligent medical and health robot in this solution utilizes a multimodal fusion technology combining voice signal analysis, text semantic understanding, and conversation context modeling to achieve real-time perception of the patient's call status and drive dynamic adjustments to speech strategies during the interaction process. Specific technical features are as follows: (41) Speech emotion recognition model training: Based on the Transformer framework, speech is converted into text tokens, and temporal information is injected through positional encoding, and a multi-layer encoder is used to extract global features. A private Conformer X multimodal attention chronic disease speech emotion recognition model is trained through CNN+Transformer to achieve recognition of the six basic emotions of speech, such as happiness, sadness, anger, fear, disgust, and surprise, and map them to emotional scene abstractions in the interaction process of medical and health intelligent robots, such as impatience, calmness, dependence, anxiety, and trust. (42) Real-time call status perception: During a call, voice stream data and semantic text stream data are collected synchronously, and multi-dimensional status features are extracted through the following submodules: (421) Speech signal feature extraction: Based on the acoustic feature extraction model (Conformer X), time domain features (such as speaking rate, intonation, volume change rate), frequency domain features (such as formant frequency, fundamental frequency jitter) and non-stationary features (such as number of pauses, choking sound detection) are extracted from the original speech to form a speech emotion feature vector; (422) Text semantic sentiment analysis: fine-tuning the real-time transcribed conversation text to output sentiment categories (impatience, calmness, dependence, trust, and 0-1 score), (e.g., “I’ve said it three times!” → impatience; “Will my blood sugar suddenly rise?” → anxiety; “Taking medicine every day is too troublesome…” → resistance); (423) Conversational context modeling: Using two-way conversations, combined with the text and state features of the past 3-5 rounds of conversations, we construct the emotion vector recognized by Conformer X in the conversational context to capture the patient's emotional trends (e.g., from calm to anxious) and potential needs (e.g., repeated mentions of "insomnia" may indicate psychological problems); (43) Real-time speech generation: Personalize the selected speech template (such as substituting the patient's name and medical history), and achieve more natural voice interaction and accurate health data collection through emotional TTS speech synthesis and context-aware slot extraction technology.

[0026] Existing TTS synthesis technologies typically only support standardized speech output and cannot adjust the tone based on the conversational context. Slot extraction often relies on rule matching, making it difficult to handle ambiguous expressions or contextual information. This solution uses emotional TTS synthesis and context-aware slot extraction technology to achieve more natural voice interaction and accurate health data collection. Specific technical features are as follows: For emotional TTS speech synthesis (speech simulation), the details are as follows: (i) Training a private voice model based on community doctors' voices: Trust between doctors and patients comes from continuous communication, but doctors have limited time. Intelligent medical and health robots, combined with TTS voice simulation, can not only relieve doctors' stress but also improve the success rate of telephone visits. However, at the beginning of the call, the robot will remind the patient that this is a voice robot telephone visit. (ii) Physician authorization and voice cloning and personalized adaptation: During the patient's first interaction, a 5-10 second standard voice clip (e.g., "Hello, I am your health care robot assistant") is collected. Voiceprint features are extracted using a speaker embedding model (e.g., d-vector or ResNet-Speaker). This is then combined with a general TTS model (e.g., FastSpeech 2) to generate a patient-specific "emotional voice cloning model," making the synthesized voice closer to the patient's familiar voice style. (iii) Ensuring the coherence of multiple rounds of dialogue: The dialogue state tracking (DST) module records the goal of the current dialogue round (e.g., “Blood pressure has been asked, and medication is pending confirmation”) and automatically adds conjunctions during TTS synthesis (e.g., “Next, I would like to know what medications you are taking recently?”) to ensure smooth dialogue.

[0027] For context-aware slot extraction, after a call ends (or during an interaction), deep semantic analysis of historical voice interaction content is performed to extract key health information and convert it into structured data. The technical implementation includes: (A) Full-text conversation cleaning: Automatic speech recognition (ASR) transcribed text is denoised (e.g., filtering out garbled text caused by coughs and environmental noise), and sentence segmentation correction (e.g., merging short sentences separated by pauses) is performed to generate standardized conversation text. (B) Multi-granularity slot extraction: A joint entity recognition and relation extraction model based on SpanBERT is used to extract health slot information of the following dimensions from the text: Basic information slot definition: age, gender, contact information (patient authorization required); Symptom description slots: symptom type (dizziness / leg pain, etc.), duration ("started three days ago"), severity ("occasionally" / "daily"); Physiological indicator slots: blood pressure ("systolic 130, diastolic 85"), pulse ("80 beats per minute"), blood sugar ("fasting 6.2"); Medication information slots: drug name ("Baixintong"), dosage ("one tablet each time"), frequency of use ("twice a day"), and medication compliance ("missed two doses recently"); (C) The model uses a contextual attention mechanism (such as the Transformer's Cross-Attention layer) to solve the problem of cross-round slot association (for example, if a patient says "I took my blood pressure yesterday" in the previous round and "high blood pressure 140" in the next round, the model needs to associate it with "blood pressure: 140 / 90 (yesterday)"); (D) Follow-up and completion of ambiguous information: If a key slot is missing (e.g., blood pressure value is not mentioned) or the expression is ambiguous (e.g., "blood pressure is a bit high"), the system will automatically trigger a follow-up question during the interaction (e.g., "Could you tell me the specific high and low blood pressure values ​​you just measured?") and mark the slot as "unconfirmed" at the end of the conversation, pushing it to the manual review module; (E) Structured data storage: The extracted slot information is converted into structured fields (e.g., "blood pressure" corresponds to "blood_pressure_systolic" and "blood_pressure_diastolic") through standardized mapping (e.g., the SNOMED CT medical terminology library) and written into the patient's electronic health record (EHR) through the HL7 FHIR interface to ensure data traceability and analyzability.

[0028] This invention combines GIS interaction technology with medical health tags. Based on the distance between the jurisdiction and offline designated activity points, it demarcates locations to achieve precise patient access and services nearby. This allows for precise management and services for patients with chronic diseases, assisting in the optimization of corresponding medical resources and the prevention and supervision of dynamic changes in the health of residents in the jurisdiction. Furthermore, it adjusts speech strategies based on the real-time status of the call (such as patient emotion recognition), combining TTS simulation and slot reading technology to make the communication process more humane. At the end of the call, the medical health intelligent robot will read the key information slots of the interactive voice content, convert it into structured data, and store it in the health record data. For example, voice features such as blood pressure, pulse, dizziness, and leg pain are converted into structured health data and stored in the corresponding patient's health record.

[0029] As for TTS simulation and slot reading technology, it has the following innovations: Multimodal emotion perception and dynamic speech adjustment: Breaking through the limitations of traditional keyword matching, through the multimodal fusion of voice + text + context, it achieves fine-grained perception of patient emotions (such as distinguishing between "anxiety" and "anger"), and drives the dynamic optimization of speech strategies, significantly improving the humanization of interactions (according to user surveys, patient satisfaction is 35% higher than that of traditional systems).

[0030] Emotional TTS and contextual slot extraction: Through voice cloning technology and emotional parameter adjustment, voice output can be tailored to each individual. Combined with pre-trained models such as SpanBERT, this solves the problem of cross-round slot association, achieving a 92% accuracy rate in extracting key health information (a 20% increase over traditional rule matching), and increasing the automation rate of structured data from 60% to 85%.

[0031] In summary, the present invention has the following advantages: Improve efficiency: Automated outbound calls cover more than 90% of standardized service scenarios, with a 100% patient consultation connection rate. Traditional manual seats will lose 40% of visiting users, freeing up 80% of manual follow-up workload.

[0032] Data continuity: Due to differences in hospital expertise and patient preferences, patients may go to Hospital A for condition A, Hospital B for condition B, and social health checkup center C for a comprehensive physical examination every year due to different circumstances at different times. Medical institution data is stored independently, and the data reported to the National Health Commission and medical insurance is also mainly from hospitals. This solution integrates data across institutions with specific users as the center through user authorization and public data authorization from the National Health Commission. It is also compatible with the upload and identification of reports from non-health and private medical examination institutions, and stores them in the user's health record. With the user's authorization, it is handed over to the contracted community health doctor or family doctor, realizing the remaining cross-institutional data and continuous tracking storage, providing doctors with a panoramic health view.

[0033] Precision service: Targeted outreach based on patients' geographic location and tags increases follow-up response rate by more than 30%.

[0034] Reduce costs: Compared with the traditional 24-hour scheduling model of human agents, the intelligent voice interaction robot system reduces labor costs by 40% and can answer repetitive questions without getting tired.

[0035] The intelligent voice robot supports interruptions midway, multi-discussion conversations, and switching topics. As long as the topic is in the RAG knowledge base, it can interact with non-task dialogues.

Claims

1. A GIS-based visualization and intelligent auxiliary management method for chronic disease patients, characterized by: The specific steps include: (1) GIS visualization of the distribution of chronic disease patients: extracting address data from electronic health case data, projecting the disease onto a GIS map based on the longitude and latitude of the community, and generating a heat map of disease aggregation; (2) Hybrid screening engine: Combine GIS interactive tools with medical and health tags, and use a privacy computing engine to ensure data security. Through public data authorization operations, source data is stored in the secure space of the National Health Commission, and different granularity data are displayed on the terminal according to the authority of the medical institution. (3) Task Configuration Center: The medical and health intelligent robot is bound to speech templates, pre-sets trigger conditions, generates personalized responses, and supports multi-round dialogue and intent recognition; (4) Macro-management analysis: Community health centers obtain regular updates of the health and medical commission data within their jurisdiction through public data authorization, or after the community health center obtains user authorization, the user regularly fills in the health status and physical examination data through the designated APP or mini-program. It also supports the health visit of medical health intelligent robots or the import of data from intelligent health consultation models. That is, after the medical health intelligent robot or model completes the health consultation for the patient, it automatically extracts data related to health, physical examination, and biochemical indicators and updates them to the corresponding user's health file. The GIS visualization platform reads the latest data updates and displays the dynamic monitoring of chronic disease patients in the jurisdiction in real time.

2. According to claim 1, a method for GIS visualization and intelligent auxiliary management of chronic disease patients is characterized in that In step (1), the specific steps are as follows: desensitize the information of chronic disease patients, retain the characteristics of symptoms and severity, extract the address data from the electronic health case data, take the value of the community without the house number, project it onto the GIS map according to the longitude and latitude of the community, and when missing values ​​are encountered, authorize and register the updated address with the patient, display different colors according to the severity of the symptoms, and generate a heat map of the concentration of different symptoms.

3. The method for GIS visualization and intelligent auxiliary management of chronic disease patients according to claim 1 is characterized in that: In step (2), specifically: select patients within the target range through the GIS interactive tool, support the standard five-level administrative boundaries of province, city, district, county, and street, and also support customized boundary fences, select one or more communities or isochronous fences, select different granularity data according to different scenarios, and cross-screen according to the patient's medical and health tags within the target fence.

4. A method for GIS visualization and intelligent auxiliary management of chronic disease patients according to claim 1 or 3, characterized in that For medical and health labels, the specific operations are as follows: (a) In the electronic medical record, the user's address is a location tag, and patients within the target service area are circled on the GIS map. From the perspective of health prevention and control, the density of patients with different diseases in different areas can be seen; (b) setting disease labels based on common chronic disease types; (c) Based on national medical and health standards or customized conditions for special populations, high, medium-high, medium and low risk levels are established for the above-mentioned different diseases, and multi-dimensional management and services are carried out based on regional location, disease symptoms and severity.

5. A method for GIS visualization and intelligent auxiliary management of chronic disease patients according to claim 1 or 3, characterized in that: For GIS interaction, GIS interaction tools refer to a series of interactive controls, operating components, or functional modules embedded in the GIS map interface. These tools enable users to deploy them in the geographic information visualization interface and support multi-dimensional interactive operations to achieve spatial positioning and attribute screening of target populations. Multi-dimensional interaction includes administrative area selection, custom fence drawing, multi-level screening, and real-time changes in map layers based on selections. This is equivalent to screening users who meet the conditions within the offline service scope. The specific operations are as follows: (21) The user opens the GIS map interface; (22) Open the custom fence in the toolbar and draw a geographic range by hand; (23) Select all patients whose addresses in the electronic health records are within the target geographic fence according to the geographic scope, and automatically extract patients who are both within the target service geographic scope and meet the target disease conditions by screening medical and health tags; (24) At this time, the GIS map displays the thermal distribution of patients who meet the conditions in real time, and the data statistics of patients who meet the conditions within the geographic fence are drawn; (25) Set the following tasks for the selected patient population.

6. The method for GIS visualization and intelligent auxiliary management of chronic disease patients according to claim 1 is characterized in that In step (4), for the medical health intelligent robot: the community health doctor authorized by the user checks and updates the health information of the corresponding user, selects the target area through the GIS visualization module according to the doctor's jurisdiction area and generates an intelligent voice access task, realizes automatic screening outbound call tasks by area, disease, severity, and special care, adjusts the speech strategy according to the real-time status of the call, combines TTS simulation and slot reading technology to make the communication process more humane, and assists the doctor to update the health information of the contracted patient. After the call ends, the medical health intelligent robot will read the key information slot of the interactive voice content, automatically structure the inquiry result data, store it in the health file data and update the health file of the corresponding patient.

7. The method for GIS visualization and intelligent auxiliary management of chronic disease patients according to claim 6 is characterized in that: During the interaction process, the medical health intelligent robot uses multimodal fusion technology of voice signal analysis, text semantic understanding, and conversation context modeling to achieve real-time perception of the patient's call status and drive dynamic adjustment of conversation strategies. The specific technical features are as follows: (41) Speech emotion recognition model training: Based on the Transformer framework training, speech is converted into text tokens, temporal information is injected through positional encoding, and global features are extracted by multi-layer encoders; a private Conformer X multimodal attention chronic disease speech emotion recognition model is trained through CNN+Transformer to achieve recognition of the six basic emotions of speech and map them to the emotional scene abstraction during the interaction process of medical and health intelligent robots; (42) Real-time call status perception: During a call, voice stream data and semantic text stream data are collected synchronously, and multi-dimensional status features are extracted through the following submodules: (421) Speech signal feature extraction: Based on the acoustic feature extraction model, time domain features, frequency domain features and non-stationary features are extracted from the original speech to form a speech emotion feature vector; (422) Text semantic sentiment analysis: fine-tuning the real-time transcribed conversation text and outputting sentiment categories; (423) Conversational context modeling: Using two-way conversations, combined with the text and state features of the past 3-5 rounds of conversations, we construct the emotion vector recognized by Conformer X in the conversational context to capture the trend of patients’ emotional changes and potential needs; (43) Real-time speech generation: Personalize the selected speech templates and achieve more natural voice interaction and accurate health data collection through emotional TTS speech synthesis and context-aware slot extraction technology.

8. The method for GIS visualization and intelligent auxiliary management of chronic disease patients according to claim 7 is characterized in that In step (43), for emotional TTS speech synthesis, the details are as follows: (i) Training a private voice model based on community doctor voices: The medical health intelligent robot combines TTS voice to simulate the doctor's voice; (ii) Physician authorization and voice cloning and personalized adaptation: During the patient's first interaction, a 5-10 second standard voice clip is collected. Voiceprint features are extracted using a speaker embedding model and combined with a general TTS model to generate a patient-specific emotional voice cloning model, making the synthesized voice closer to the patient's familiar voice style. (iii) Ensuring multi-round dialogue coherence: The dialogue state tracking module records the goals of the current dialogue turn and automatically adds connectives during TTS synthesis to ensure smooth dialogue.

9. The method for GIS visualization and intelligent auxiliary management of chronic disease patients according to claim 7 is characterized in that In step (43), for context-aware slot extraction, after the call ends or during the interaction, the historical voice interaction content is deeply semantically parsed to extract key health information and convert it into structured data; Technical implementation includes: (A) Full conversation text cleaning: De-noise and correct sentence segments of the ASR transcribed text to generate standardized conversation text; (B) Multi-granularity slot extraction: A joint entity recognition and relation extraction model based on SpanBERT is used to extract health slot information of the following dimensions from the text: basic information slot, symptom description slot, physiological indicator slot, and medication information slot; (C) The model solves the cross-round slot association problem through the contextual attention mechanism; (D) Follow-up and completion of ambiguous information: If a key slot is detected to be missing or vaguely expressed, the system will automatically trigger follow-up questions during the interaction and mark the slot as "unconfirmed" after the conversation ends, pushing it to the manual review module; (E) Structured data storage: The extracted slot information is converted into structured fields through standardized mapping and written into the patient's electronic health record through the HL7 FHIR interface to ensure that the data is traceable and analyzable.

Images