An AI-based family doctor management system for gestational diabetes patients

Abstract

This invention relates to the field of medical management system technology and discloses an artificial intelligence-based family doctor management system for gestational diabetes patients. The system includes a data acquisition module, a data preprocessing module, an artificial intelligence analysis and early warning module, a personalized intervention module, an interactive display module, and a data storage module. These modules are interconnected via an encrypted communication link. By integrating data acquisition, artificial intelligence analysis, and personalized intervention modules, this invention achieves precise, end-to-end management of the patient's condition. The system can collect multi-dimensional data on the patient's blood glucose, diet, and exercise in real time. After preprocessing, the data is used to conduct risk warnings and condition assessments through an AI model. Compared to traditional family management models, this significantly improves the timeliness and accuracy of condition monitoring, enabling early prediction of abnormal risks and sending reminders to help patients and doctors respond quickly and reduce the risk of complications from gestational diabetes.

Description

Technical Field

[0001] This invention relates to the field of medical management system technology, specifically to an artificial intelligence-based family doctor management system for gestational diabetes patients. Background Technology

[0002] Gestational diabetes mellitus (GDM) refers to glucose intolerance that first occurs or is discovered during pregnancy. Its global incidence has been rising steadily in recent years, reaching 10%-20% in my country. This condition not only increases the risk of preeclampsia and postpartum type 2 diabetes in pregnant women, but also easily leads to adverse perinatal outcomes such as macrosomia and neonatal hypoglycemia. Because GDM lasts throughout the second and third trimesters, patients need to complete blood glucose monitoring and lifestyle interventions outside the hospital for the vast majority of the time. Traditional outpatient follow-up models struggle to provide real-time feedback, making home self-management a key bottleneck affecting blood glucose control rates.

[0003] Current gestational diabetes management technologies primarily rely on mobile health applications and remote monitoring platforms, but these have certain limitations: most systems only collect blood glucose data via Bluetooth and display it in charts, lacking dynamic blood glucose prediction models tailored to the physiological characteristics of pregnancy (such as gestational age progression and hormonal fluctuations), and are unable to provide early warnings for specific risk trends such as nocturnal hypoglycemia or postprandial hyperglycemia drift. Therefore, there is an urgent need to develop an intelligent family doctor management system with proactive learning capabilities, capable of risk warnings and personalized adaptive solutions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based family doctor management system for gestational diabetes patients. This system solves the pain points of insufficient professionalism and weak intervention targeting in existing family management of gestational diabetes. By integrating data collection, artificial intelligence analysis, and personalized intervention modules, it achieves precise control over the entire process of the patient's condition and solves the aforementioned technical problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a family doctor management system for gestational diabetes patients based on artificial intelligence, comprising a data acquisition module, a data preprocessing module, an artificial intelligence analysis and early warning module, a personalized intervention module, an interactive display module, and a data storage module, wherein each module is interconnected through an encrypted communication link; The data acquisition module is used to collect multi-dimensional dynamic data of patients, including blood glucose monitoring data (fasting, 1h / 2h postprandial, random values ​​at night and timestamps), physiological characteristics data during pregnancy (gestational week, fluctuations of hormones such as estradiol, fundal height and abdominal circumference), lifestyle data and basic health data, and supports automatic expansion of physiological characteristic collection dimensions according to gestational week. The data preprocessing module has a built-in abnormal data identification algorithm specifically for gestational diabetes. It performs layered processing on the multi-dimensional raw data collected by the data acquisition module, converting data of different dimensions and units into a unified format and outputting a standardized dataset that can be directly used for model calculation. The artificial intelligence analysis and early warning module incorporates a dynamic blood glucose prediction model trained jointly on clinical cases of gestational diabetes and physiological characteristics during pregnancy. This dynamic blood glucose prediction model introduces a differentiated blood glucose risk early warning calculation formula: in: This is a blood sugar risk warning index. , and To adapt to the dynamic weighting coefficients during pregnancy and + + =1, This is the current blood glucose level. To correspond to the normal blood glucose threshold for the gestational week, To correspond to the standard deviation of blood glucose during gestational week, The change in blood glucose over T time units (T=30min) is the amount of change in blood glucose. This is a correction factor for the effects of hormones. The model takes the preprocessed standardized dataset as input and combines the patient's gestational age progression and hormone fluctuation patterns to achieve autonomous iterative optimization. It accurately identifies nocturnal hypoglycemia (<3.3mmol / L) and postprandial hyperglycemia drift (2h after a meal >8.5mmol / L and lasting ≥30min), and outputs mild, moderate and severe graded warnings and blood glucose prediction curves 30-60min in advance, which are simultaneously sent to the patient and doctor. The personalized intervention module, based on the prediction results and risk level of the artificial intelligence analysis and early warning module, combined with the patient's gestational age, hormonal characteristics, and dietary and exercise habits, adaptively generates a dynamic personalized management plan, including dietary recommendations that fit the gestational age, exercise recommendations that are appropriate for pregnancy contraindications, and blood glucose monitoring frequency recommendations based on the risk level. The interactive display module is compatible with both the patient and doctor ends. The patient end displays blood glucose data, warning prompts, and intervention plans, and supports feedback and questions. The doctor end can view the patient's health data and warning information, and can manually adjust the intervention plan and synchronize it. The data storage module uses encrypted storage to save various patient data, early warning records, and intervention plans. It supports retrieval by gestational week and risk level, providing support for model iteration and clinical research.

[0006] Preferably, the data acquisition module adopts a mode combining Bluetooth 5.2 real-time transmission and touch-based manual supplementation and input. Blood glucose monitoring data is collected wirelessly with a home blood glucose meter via Bluetooth, while pregnancy hormone fluctuation data and fundal height and abdominal circumference data are supplemented by the user through a touch input interface. The input interface is adapted to the operating habits of pregnant users and supports voice-assisted input.

[0007] Preferably, the gestational diabetes-specific abnormal data identification algorithm built into the data preprocessing module uses a combination of interquartile range and local anomaly factor method to remove outliers. Specifically, key data such as blood glucose and hormones are identified as abnormal using the interquartile range method, auxiliary data such as lifestyle data are identified as abnormal using the local anomaly factor method, and missing data are supplemented using linear interpolation.

[0008] Preferably, the dynamic blood glucose prediction model in the artificial intelligence analysis and early warning module is constructed using a fusion algorithm of CNN and Transformer. The CNN network extracts local features of the patient's blood glucose and hormone data, and the Transformer network captures the long-term dependence between gestational age progression and blood glucose changes. At the same time, a physiological characteristic constraint layer during pregnancy is embedded, and the extracted local features, long-term dependent features, and gestational age and hormone fluctuation parameters are substituted into the blood glucose risk early warning calculation formula.

[0009] Preferably, in the differential blood glucose risk warning calculation formula, the dynamic weighting coefficient , and Adjustments are made dynamically according to the patient's gestational age, during the first trimester. =0.3、 =0.3、 =0.4, mid-pregnancy =0.4、 =0.3、 =0.3, late pregnancy =0.3、 =0.4、 =0.3, Hormone Influence Correction Factor The value range is 0.8-1.2.

[0010] Preferably, the AI ​​analysis and early warning module provides a mild, moderate, and severe risk level warning based on a blood glucose risk warning index. The criteria for judgment are as follows: when A value <0.3 indicates a mild warning, and a text notification will be sent only to the patient. When 0.3≤ A reading of <0.7 indicates a moderate warning, and text and voice prompts are simultaneously sent to the patient's device. when A value ≥0.7 indicates a severe warning, triggering a text and voice alert to the patient's device, along with an additional emergency alert to the doctor's device, and marking the patient as a priority.

[0011] Preferably, the dietary recommendations generated by the personalized intervention module combine the patient's nutritional needs during pregnancy and blood glucose control targets, specifically including daily calorie intake thresholds, carbohydrate / protein / fat ratios, and specific meal examples suitable for pregnancy, while also indicating prohibited and alternative ingredients.

[0012] Preferably, the patient-side of the interactive display module supports the function of exporting blood glucose data, which is used to export blood glucose monitoring records, early warning records and intervention plan execution data in Excel format. The doctor-side supports batch viewing of early warning statistics data of patients under their jurisdiction and generating weekly / monthly early warning summary reports.

[0013] Preferably, the data storage module uses an encryption algorithm combined with pregnancy privacy desensitization processing to achieve encrypted data storage. During the desensitization process, only the patient's gestational age and blood glucose-related core diagnostic and treatment data are retained, while sensitive information such as name and ID number are hidden. At the same time, a dual backup mechanism is set up, which automatically completes local + cloud dual backup every morning. The backup data is retained for 90 days, supports accidental deletion recovery, and verifies data integrity in real time during the backup process.

[0014] Preferably, the doctor's terminal has a built-in integrated function of intelligent follow-up reminder and follow-up effect evaluation. In addition to setting differentiated follow-up cycles according to blood glucose risk level (every 2 weeks for mild cases, weekly for moderate cases, and every 3 days for severe cases), it can also automatically capture the patient's blood glucose control data and intervention plan implementation during the follow-up period, generate a follow-up effect score, and adaptively adjust the subsequent follow-up cycle and intervention plan recommendations based on the score. At the same time, it supports doctors to manually mark the key points of the follow-up, and the marked content is archived in association with the patient's blood glucose data.

[0015] Compared with existing technologies, this invention provides an artificial intelligence-based family doctor management system for gestational diabetes patients, which has the following beneficial effects: 1. This invention integrates data collection, artificial intelligence analysis, and personalized intervention modules to achieve precise control over the entire process of a patient's condition. The system can collect multi-dimensional data such as blood glucose, diet, and exercise in real time. After preprocessing, the AI ​​model is used for risk warning and condition assessment. Compared with the traditional family management model, it significantly improves the timeliness and accuracy of condition monitoring, can predict abnormal risks in advance and push reminders, helps patients and doctors respond quickly, reduces the risk of complications of gestational diabetes, and reduces the burden of daily follow-up visits for doctors. It makes family management more professional and scientific, and provides reliable protection for the health of patients during pregnancy.

[0016] 2. This invention presents patients with concise and easy-to-understand medical reports and intervention plans through an interactive display module, catering to the needs of pregnant patients. It also allows doctors to view patient data in real time and adjust intervention strategies, enabling precise delivery of personalized dietary, exercise, and medication guidance. Compared to conventional management methods, this invention breaks down the time and space limitations of doctor-patient communication, allowing patients to receive professional medical guidance without frequent hospital visits, reducing medical costs. Furthermore, through data storage and model iteration, it continuously optimizes intervention plans to adapt to changes in the patient's condition at different stages of pregnancy, helping patients maintain stable blood sugar control and ensuring the health of both mother and child. Attached Figure Description

[0017] Figure 1 This is a logic block diagram showing the relationships between the various modules of this invention; Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example: Please see Figure 1 The present invention provides an artificial intelligence-based family doctor management system for patients with gestational diabetes, including a data acquisition module, a data preprocessing module, an artificial intelligence analysis and early warning module, a personalized intervention module, an interactive display module, and a data storage module. The modules are interconnected through an encrypted communication link. The data acquisition module is used to collect multi-dimensional dynamic data from patients. It adopts a hybrid acquisition mode of Bluetooth 5.2 real-time transmission, touch-screen manual supplementary input, and voice-assisted input to adapt to the operating habits of pregnant patients, achieving comprehensive and convenient acquisition of multi-dimensional dynamic data. It also supports automatic expansion of physiological characteristic acquisition dimensions according to gestational week. The specific acquisition content and implementation method are as follows: 1. Automatic data collection: Blood glucose monitoring data (fasting, 1h / 2h postprandial, random values ​​at night and corresponding timestamps) are wirelessly linked with the home blood glucose meter via Bluetooth 5.2. After the blood glucose meter completes the test, the data is synchronized to the system in real time without the need for manual entry by the patient. The synchronization delay is ≤1s, ensuring the timeliness of the data. 2. Manual + Voice Data Entry: Patients can complete data entry for pregnancy physiological characteristics (gestational age, estradiol and other hormone fluctuations, fundal height and abdominal circumference measurements), lifestyle data (daily food types and intake, exercise duration / type, sleep patterns), and basic health data (blood pressure, weight, heart rate) through the system's touch input interface. The interface is designed with large buttons and simple interactive logic, taking into account the expanding abdomen during pregnancy. It also supports voice-assisted entry; patients can speak the measured values, and the system automatically recognizes and fills them into the corresponding fields with an accuracy rate ≥95%. 3. Adaptive gestational age data collection: The system automatically calculates gestational age based on the patient's last menstrual period. In the first trimester (1-12 weeks), only core blood glucose, basic health, and basic hormone data are collected. In the second trimester (13-28 weeks), data collection dimensions such as fundal height, abdominal circumference, and progesterone fluctuations are automatically added. In the third trimester (29-40 weeks), data collection dimensions such as fetal movement monitoring and daily weight changes are further expanded to achieve accurate matching between the data collection dimensions and the physiological stages of pregnancy.

[0020] The data preprocessing module incorporates a built-in abnormal data identification algorithm specific to gestational diabetes. It performs stratified processing on the collected multi-dimensional raw data, ultimately outputting a standardized dataset to provide high-quality data support for subsequent model calculations. The specific processing flow is as follows: 1. Outlier Identification and Removal: A combination of interquartile range (IVR) and local outlier factor (LOF) methods is used for identification. For key diagnostic data related to hormones such as blood glucose and estradiol, IVR is used to calculate upper and lower thresholds. Values ​​exceeding the thresholds are identified as outliers and removed directly. For auxiliary data such as lifestyle and sleep patterns, IVR is used to calculate local outlier factors for data points. Factor values ​​greater than a preset threshold (set to 1.5) are identified as outliers and removed to avoid interference from outlier data in the model analysis. 2. Missing Data Supplementation: For missing data caused by equipment failure or missed patient entries during the collection process, linear interpolation is used to supplement the missing data. Based on the three valid data points before and after the missing data point, the missing value is calculated through linear fitting to ensure the integrity of the dataset. For more than five consecutive missing data points, the system will push a supplementary entry reminder to the patient in the interactive display module. 3. Data standardization processing: Convert raw data of different dimensions and units into a unified format, specifically: blood glucose data are uniformly converted to mmol / L, weight is uniformly converted to kg, fundal height / abdominal circumference is uniformly converted to cm, and time is uniformly converted to 24-hour format. At the same time, all data are normalized (mapped to the [0,1] interval), and finally a structured and standardized dataset is output, which can be directly input into the artificial intelligence analysis and early warning module for calculation.

[0021] Artificial intelligence analysis and early warning: This module is the core processing module of the system. It incorporates a dynamic blood glucose prediction model jointly trained based on clinical cases of gestational diabetes and physiological characteristics of pregnancy. The model is constructed using a fusion algorithm of CNN and Transformer, and embeds a constraint layer of physiological characteristics of pregnancy. Combined with a differentiated blood glucose risk early warning calculation formula, it achieves accurate graded early warning of blood glucose risk and prediction of blood glucose trends. Specifically, it includes three core links: model construction, risk index calculation, and graded early warning. 1. Construction and training of dynamic blood glucose prediction model; Feature extraction: Local features of patients' blood glucose and hormone data are extracted using CNN networks (such as a single blood glucose spike / drop and short-term hormone fluctuations). The self-attention mechanism of Transformer networks is used to capture the long-term dependence of gestational age progression on blood glucose changes (such as the overall upward trend of blood glucose in late pregnancy as gestational age increases). Physiological constraints: An embedded layer of physiological characteristics of pregnancy is used as a constraint, incorporating pregnancy-specific features such as gestational age, hormone fluctuation amplitude, fundal height and abdominal circumference changes as constraints into the feature extraction results, thus avoiding prediction results that deviate from the physiological laws of pregnancy. Model training: Using multidimensional data from 10,000+ clinical cases of gestational diabetes as the training set and a preprocessed standardized dataset as input, the model was iteratively trained using the gradient descent method. During the training process, real-time patient data was incorporated to achieve autonomous iterative optimization and continuously improve prediction accuracy. The model's blood glucose prediction error was ≤0.3mmol / L.

[0022] 2. Calculation of differentiated blood glucose risk warning index; The model substitutes standardized datasets into a differentiated blood glucose risk warning formula to calculate a real-time blood glucose risk warning index for patients. The calculation formula is: The parameter settings and calculation rules are as follows: Dynamic weighting coefficients: , and Adjustments are made dynamically according to gestational age, and the following conditions are met. + + =1; Early pregnancy =0.3、 =0.3、 =0.4, focusing on the impact of hormonal fluctuations on blood sugar; mid-pregnancy =0.4、 =0.3、 =0.3, focusing on the deviation of the current blood glucose level from the normal threshold; late pregnancy =0.3、 =0.4、 =0.3, focusing on short-term blood glucose trends; Basic parameters: This is the patient's current blood glucose level. This refers to the normal blood glucose threshold for the corresponding gestational week (e.g., the normal fasting blood glucose threshold in late pregnancy is 3.3-5.6 mmol / L). The standard deviation of blood glucose at corresponding gestational weeks (derived from clinical big data statistics). This represents the change in blood glucose levels over 30 minutes. This refers to the amplitude of hormone fluctuations; Correction factor: Hormone effect correction factor The value ranges from 0.8 to 1.2, and doctors can manually fine-tune it on the doctor's end according to the individual hormone fluctuations of the patient.

[0023] 3. Tiered early warning and trend forecasting; The model uses a blood glucose risk warning index To establish a tiered early warning system for mild, moderate, and severe cases, and to output blood glucose prediction curves 30-60 minutes in advance, the warning information is synchronized to both the patient's and doctor's devices. The specific warning rules are as follows: Mild warning (R<0.3): Only a text prompt is pushed to the patient's end, such as "The current blood sugar risk is low, and it is recommended to continue to maintain the current diet and exercise habits", while displaying the blood sugar prediction curve; Moderate warning (0.3≤R<0.7): Text and voice prompts are pushed to the patient's device simultaneously, such as "The current blood sugar is fluctuating abnormally. Postprandial hyperglycemia is expected in 30 minutes. It is recommended to reduce the intake of refined carbohydrates." Severe Warning (R≥0.7): The patient receives a text and voice emergency alert, while the doctor receives an emergency alert with the patient's priority level marked (highest). The doctor receives a pop-up notification displaying the patient's real-time blood glucose data, gestational age, and medical history, facilitating timely intervention. This module accurately identifies high-risk conditions for gestational diabetes, such as nocturnal hypoglycemia (<3.3mmol / L) and postprandial hyperglycemia drift (2h postprandial >8.5mmol / L lasting ≥30min). For nocturnal hypoglycemia, the system will send a vibration and voice alert to the patient 60 minutes before the predicted hypoglycemia occurs, preventing accidents caused by nighttime hypoglycemia.

[0024] The personalized intervention module, based on the blood glucose prediction results and risk level from the AI-powered analysis and early warning module, combined with the patient's gestational age, hormonal characteristics, dietary and exercise habits, and pregnancy contraindications, adaptively generates a dynamic and personalized management plan. The plan is updated in real time as the patient's blood glucose levels change and pregnancy progresses, and includes three core components: dietary recommendations, exercise recommendations, and recommendations for blood glucose monitoring frequency. 1. Dietary Recommendations: Based on the patient's nutritional needs during pregnancy and blood glucose control targets, accurately calculate the daily calorie intake thresholds (1800-2000 kcal in early pregnancy, 2100-2300 kcal in mid-pregnancy, and 2300-2500 kcal in late pregnancy, adjusted appropriately according to the patient's body mass index), determine the carbohydrate / protein / fat ratio (carbohydrates 50%-55%, protein 20%-25%, fat 20%-25%), and provide specific meal examples suitable for pregnancy (such as mixed grain rice, steamed fish, and cold vegetable salad), and indicate foods prohibited during pregnancy (such as raw seafood and high-sugar pastries) and alternative foods (such as using xylitol instead of sucrose and buckwheat noodles instead of refined noodles). For patients with moderate or higher blood glucose warnings, automatically reduce the carbohydrate ratio and increase dietary fiber recommendations. 2. Exercise recommendations: Adapt to pregnancy contraindications and avoid strenuous exercise and exercises that put pressure on the abdomen. Develop personalized exercise plans for patients with different gestational weeks and different blood sugar risk levels. For example, in early pregnancy, slow walking and prenatal yoga (20-30 minutes daily) are recommended. In mid-to-late pregnancy, walking and pelvic floor muscle training (30-40 minutes daily) are recommended. For patients with severe blood sugar warnings, it is recommended to engage in light exercise under the guidance of a doctor to avoid blood sugar fluctuations caused by self-exercise. 3. Recommended frequency of blood glucose monitoring: Adjust dynamically according to risk level. Patients with mild risk warning are advised to monitor 4 times a day (fasting + 2 hours after each meal), patients with moderate risk warning are advised to monitor 7 times a day (fasting + 1 / 2 hours after each meal + random monitoring at night), and patients with severe risk warning are advised to increase the frequency of nighttime monitoring (every 3-4 hours) under the guidance of a doctor.

[0025] The interactive display module is compatible with both patient-side (APP / Mini Program) and doctor-side (web page / PC client) terminals, enabling bidirectional collaboration in data display, information interaction, and command issuance. Each terminal has its own functional focus, and data is synchronized in real time. 1. Patient-side functions: Displays real-time blood glucose data, historical blood glucose trend curves, blood glucose risk warnings, and personalized intervention plans. It supports patient feedback (such as inquiries about dietary advice or requests for help with abnormal blood glucose levels), with submitted questions pushed to the doctor's end in real time. It also supports blood glucose data export, allowing for one-click export of blood glucose monitoring records, warning records, and intervention plan execution data in Excel format, conveniently provided to doctors for reference during outpatient visits. The patient-side features a dedicated message reminder area where all warning messages, doctor replies, and follow-up reminders are centrally displayed to prevent omissions. Doctor-side functions: Doctors can view comprehensive health data, real-time alerts, and intervention plan implementation status for their managed patients. Doctors can manually adjust intervention plans based on individual patient conditions, and the adjustments are synchronized to the patient's end. The system supports batch viewing of alert statistics for managed patients and automatically generates weekly / monthly alert summary reports. These reports include core indicators such as patient alert level distribution, blood glucose target achievement rate, and intervention plan implementation rate, facilitating group management by doctors. Doctors can also set patient priority tags, marking severely alerted patients in red for priority handling.

[0026] The data storage module employs encrypted storage, privacy desensitization, and dual backup to ensure secure storage, convenient retrieval, and reliable retention of patient data, providing high-quality data support for model iteration and clinical research. The specific implementation method is as follows: 1. Encryption and desensitization storage: The AES-256 encryption algorithm is used to encrypt and store all patient data (collected data, early warning records, intervention plans, follow-up data). At the same time, privacy desensitization processing is performed during pregnancy, retaining only the patient's gestational age and blood glucose-related core diagnosis and treatment data, while hiding sensitive information such as name, ID number, and home address. Data is linked through a unique patient identification code to protect patient privacy. 2. Convenient search: Supports precise data retrieval by gestational week, risk level, warning type, etc. Doctors and researchers can filter data according to their needs, and the search results only display the core information after anonymization; 3. Dual Backup and Recovery: A dual backup mechanism is set up to automatically complete a dual backup of the local server and the cloud server at 3:00 AM every day. During the backup process, MD5 checksum is used to verify data integrity in real time to avoid data corruption. Backup data is retained for 90 days and supports recovery from accidental deletion. If a patient or doctor accidentally deletes data, they can apply for data recovery through the system within 90 days, with a recovery time of ≤24 hours.

[0027] The overall operation flow of the family doctor management system for gestational diabetes patients based on artificial intelligence according to the present invention is as follows: 1. Data collection: Patients complete multi-dimensional data collection through Bluetooth synchronization of blood glucose meter + manual / voice input. The system automatically expands the collection dimensions according to gestational week. The collected data is transmitted to the data preprocessing module through an encrypted communication link. 2. Data preprocessing: The raw data is processed by removing outliers, filling in missing values, and standardizing the data to output a standardized dataset, which is then transmitted to the artificial intelligence analysis and early warning module. 3. Intelligent Analysis and Early Warning: The dynamic blood glucose prediction model extracts and analyzes features from the standardized dataset, substitutes them into the differential blood glucose risk early warning calculation formula to obtain the risk index R, realizes the graded early warning of mild, moderate and severe conditions, outputs the blood glucose prediction curve 30-60 minutes in advance, and synchronizes the early warning information and prediction results to the personalized intervention module, interactive display module and data storage module. 4. Personalized intervention plan generation: Based on the early warning results, the patient's gestational age and individual characteristics, the personalized intervention module adaptively generates dynamic suggestions on diet, exercise and monitoring frequency, which are transmitted to the interactive display module and stored in the data storage module; 5. Dual-terminal interaction and intervention: The patient's end displays warning information and intervention plan, and the patient can provide feedback and questions; the doctor's end views patient data and warning information, can manually adjust the intervention plan and synchronize it to the patient's end, and the doctor can intervene and provide guidance in a timely manner when there is a severe warning. 6. Data storage and follow-up management: The data storage module encrypts and de-identifies all data and performs dual backups, supporting multi-dimensional retrieval; the doctor's end performs intelligent follow-up based on risk level, automatically evaluates follow-up effectiveness and adjusts follow-up cycle and intervention plan, and all follow-up data is linked and archived. 7. Model Iteration and Optimization: Based on stored real-time patient data and clinical follow-up data, the AI ​​analysis and early warning module enables autonomous iterative optimization of the model, continuously improving the accuracy of blood glucose prediction and risk warning.

[0028] This embodiment of the AI-based family doctor management system for gestational diabetes patients addresses the physiological characteristics and home management needs of gestational diabetes. It facilitates data collection, standardizes data processing, enhances the precision of analysis and early warning, personalizes intervention plans, and ensures secure data storage. It overcomes the shortcomings of traditional outpatient follow-up and existing management systems, enabling real-time and intelligent family doctor management of gestational diabetes patients throughout their entire outpatient cycle. This effectively improves the rate of patients achieving blood glucose control targets, reduces the risk of adverse perinatal outcomes, and provides a high-quality data source for clinical research on gestational diabetes.

[0029] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A family doctor management system for gestational diabetes patients based on artificial intelligence, characterized in that: It includes a data acquisition module, a data preprocessing module, an artificial intelligence analysis and early warning module, a personalized intervention module, an interactive display module, and a data storage module. All modules are interconnected through an encrypted communication link. The data acquisition module is used to collect multi-dimensional dynamic data of patients, including blood glucose monitoring data (fasting, 1h / 2h postprandial, random values ​​at night and timestamps), physiological characteristics data during pregnancy (gestational week, fluctuations of hormones such as estradiol, fundal height and abdominal circumference), lifestyle data and basic health data, and supports automatic expansion of physiological characteristic collection dimensions according to gestational week. The data preprocessing module has a built-in abnormal data identification algorithm specifically for gestational diabetes. It performs layered processing on the multi-dimensional raw data collected by the data acquisition module, converting data of different dimensions and units into a unified format and outputting a standardized dataset that can be directly used for model calculation. The artificial intelligence analysis and early warning module incorporates a dynamic blood glucose prediction model trained jointly on clinical cases of gestational diabetes and physiological characteristics during pregnancy. This dynamic blood glucose prediction model introduces a differentiated blood glucose risk early warning calculation formula: in: This is a blood sugar risk warning index. , and To adapt to the dynamic weighting coefficients during pregnancy and + + =1, This is the current blood glucose level. To correspond to the normal blood glucose threshold for the gestational week, To correspond to the standard deviation of blood glucose during gestational week, The change in blood glucose over T time units (T=30min) is the amount of change in blood glucose. This is a correction factor for the effects of hormones. The model takes the preprocessed standardized dataset as input and combines the patient's gestational age progression and hormone fluctuation patterns to achieve autonomous iterative optimization. It accurately identifies nocturnal hypoglycemia (<3.3mmol / L) and postprandial hyperglycemia drift (2h after a meal >8.5mmol / L and lasting ≥30min), and outputs mild, moderate and severe graded warnings and blood glucose prediction curves 30-60min in advance, which are simultaneously sent to the patient and doctor. The personalized intervention module, based on the prediction results and risk level of the artificial intelligence analysis and early warning module, combined with the patient's gestational age, hormonal characteristics, and dietary and exercise habits, adaptively generates a dynamic personalized management plan, including dietary recommendations that fit the gestational age, exercise recommendations that are appropriate for pregnancy contraindications, and blood glucose monitoring frequency recommendations based on the risk level. The interactive display module is compatible with both the patient and doctor ends. The patient end displays blood glucose data, warning prompts, and intervention plans, and supports feedback and questions. The doctor end can view the patient's health data and warning information, and can manually adjust the intervention plan and synchronize it. The data storage module uses encrypted storage to save various patient data, early warning records, and intervention plans. It supports retrieval by gestational week and risk level, providing support for model iteration and clinical research.

2. The family doctor management system for gestational diabetes patients based on artificial intelligence according to claim 1, characterized in that: The data acquisition module adopts a combination of Bluetooth 5.2 real-time transmission and touch-based manual input. Blood glucose monitoring data is collected wirelessly via Bluetooth with a home blood glucose meter, while pregnancy hormone fluctuation data and fundal height and abdominal circumference data are supplemented by the user through the touch input interface. The input interface is adapted to the operating habits of pregnant users and supports voice-assisted input.

3. The family doctor management system for gestational diabetes patients based on artificial intelligence according to claim 1, characterized in that: The data preprocessing module has a built-in abnormal data identification algorithm for gestational diabetes. It uses a combination of interquartile range and local anomaly factor method to remove outliers. Specifically, the interquartile range method is used to identify abnormalities in key data such as blood glucose and hormones, while the local anomaly factor method is used to identify abnormalities in lifestyle auxiliary data. Missing data is supplemented by linear interpolation.

4. The family doctor management system for gestational diabetes patients based on artificial intelligence according to claim 1, characterized in that: The dynamic blood glucose prediction model in the artificial intelligence analysis and early warning module is constructed using a fusion algorithm of CNN and Transformer. The CNN network extracts local features of the patient's blood glucose and hormone data, and the Transformer network captures the long-term dependence of gestational age progression and blood glucose changes. At the same time, a physiological characteristic constraint layer during pregnancy is embedded. The extracted local features, long-term dependent features, and gestational age and hormone fluctuation parameters are substituted into the blood glucose risk early warning calculation formula.

5. The family doctor management system for gestational diabetes patients based on artificial intelligence according to claim 1, characterized in that: In the differential blood glucose risk warning calculation formula, the dynamic weighting coefficient , and Adjustments are made dynamically according to the patient's gestational age, during the first trimester. =0.3、 =0.3、 =0.4, mid-pregnancy =0.4、 =0.3、 =0.3, late pregnancy =0.3、 =0.4、 =0.3, Hormone Influence Correction Factor The value range is 0.8-1.

2.

6. The family doctor management system for gestational diabetes patients based on artificial intelligence according to claim 1, characterized in that: The artificial intelligence analysis and early warning module provides a mild, moderate, and severe level of early warning, based on a blood glucose risk warning index. The criteria for judgment are as follows: when A value <0.3 indicates a mild warning, and a text notification will be sent only to the patient. When 0.3≤ A reading of <0.7 indicates a moderate warning, and text and voice prompts are simultaneously sent to the patient's device. when A value ≥0.7 indicates a severe warning, triggering a text and voice alert to the patient's device, along with an additional emergency alert to the doctor's device, and marking the patient as a priority.

7. The family doctor management system for gestational diabetes patients based on artificial intelligence according to claim 1, characterized in that: The personalized intervention module generates dietary recommendations that take into account the patient's nutritional needs during pregnancy and blood glucose control goals. Specifically, these recommendations include daily calorie intake thresholds, carbohydrate / protein / fat ratios, and specific meal examples suitable for pregnancy, while also indicating prohibited and alternative ingredients.

8. The family doctor management system for gestational diabetes patients based on artificial intelligence according to claim 1, characterized in that: The patient-side of the interactive display module supports the function of exporting blood glucose data, which can be used to export blood glucose monitoring records, early warning records and intervention plan execution data in Excel format. The doctor-side supports batch viewing of early warning statistics for patients under their jurisdiction and generating weekly / monthly early warning summary reports.

9. The family doctor management system for gestational diabetes patients based on artificial intelligence according to claim 1, characterized in that: The data storage module uses encryption algorithms combined with pregnancy privacy desensitization processing to achieve encrypted data storage. During the desensitization process, only the patient's gestational age and blood sugar-related core diagnostic and treatment data are retained, while sensitive information such as name and ID number are hidden. At the same time, a dual backup mechanism is set up, which automatically completes local + cloud dual backup every day at midnight. Backup data is retained for 90 days, supports accidental deletion recovery, and verifies data integrity in real time during the backup process.

10. A family doctor management system for gestational diabetes patients based on artificial intelligence according to claim 1, characterized in that: The doctor's terminal has a built-in integrated function of intelligent follow-up reminder and follow-up effect evaluation. In addition to setting differentiated follow-up cycles according to blood glucose risk level (every 2 weeks for mild, weekly for moderate, and every 3 days for severe), it can also automatically capture blood glucose control data and intervention plan implementation during the patient's follow-up period, generate follow-up effect score, and adaptively adjust subsequent follow-up cycles and intervention plan recommendations based on the score. At the same time, it supports doctors to manually mark the key points of follow-up, and the marked content is archived in association with the patient's blood glucose data.

Images