Obstetrical risk intelligent early warning system based on AI analysis

Abstract

The invention relates to the technical field of medical health information processing, in particular to an obstetric risk intelligent early warning system based on AI analysis. Comprising a data table creation module, a multi-source data acquisition module, a fusion processing module, an AI analysis module, an AI early warning module, an auditing module, a distribution module, a manual auditing module and a receiving module. By integrating electronic medical records, wearable equipment and manual input data, an omnibearing pregnant and lying-in woman health portrait is constructed, and the limitation of a single data source is overcome. Data are sorted and fused based on a time axis, evolution trends and association relationships of physiological parameters, symptoms and signs can be accurately captured, and subtle changes and early risk signals which are easy to omit in traditional discontinuous antenatal care are identified.

Description

Technical Field

[0001] This invention relates to the field of medical and health information processing technology, specifically to an AI-based intelligent early warning system for obstetric risks. Background Technology

[0002] The widespread adoption of electronic medical record systems, the extensive use of wearable medical devices, and the maturity of high-throughput gene sequencing technologies have provided unprecedented data support for obstetric clinical decision-making. Especially in the field of dynamic health monitoring, real-time physiological parameter acquisition and intelligent analysis technologies have gradually become the core driving force for improving diagnostic and treatment accuracy.

[0003] However, the limitations of existing technology systems have led to some problems in current obstetric and gynecological health management technologies. For example, mainstream electronic medical record systems mainly archive structured data and cannot continuously track the dynamic evolution of physiological parameters, resulting in a lag in risk assessment. While wearable medical devices can collect some physiological characteristic parameters of users in real time, they cannot be effectively integrated with electronic medical record systems, thus preventing the two from effectively complementing each other to play a role in risk warning. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based intelligent early warning system for obstetric risks, which solves the problem that existing technologies cannot effectively combine electronic medical records, wearable medical devices, etc., in real time to monitor and warn pregnant women.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] An AI-based intelligent early warning system for obstetric risks includes: a data table creation module for creating dynamic data tables; a multi-source data acquisition module for collecting case data from electronic medical record systems, dynamic data from wearable medical devices, and manually entered data; a fusion processing module for organizing and fusing case data, dynamic data, and manual data based on a timeline and filling them into the corresponding positions in the dynamic data table; an AI analysis module for analyzing the dynamic data table, marking risk points, risk triggers, and the duration of risk points, and assigning risk values ​​to risk points; and an AI early warning module for generating preliminary precaution reports based on risk points, risk triggers, and the duration of risk points. If a risk trigger is triggered within the duration of a risk point, a preliminary early warning report is generated, including the warning level and risk assessment criteria. The system comprises the following modules: a risk handling plan; an audit module for receiving preliminary precaution reports and preliminary warning reports, matching the preliminary precaution reports with a first manual audit plan based on a preset audit plan, and formulating a second manual audit plan for the preliminary warning reports based on the risk level; an allocation module for receiving preliminary precaution reports and their matching first manual audit plans, as well as preliminary warning reports and their matching second manual audit plans, and allocating them to corresponding auditors for manual review based on the first and second manual audit plans; a manual review module for receiving the allocated preliminary precaution reports and preliminary warning reports, and reviewing and modifying them to form final precaution reports and final warning reports; and a receiving module for receiving final precaution reports and final warning reports.

[0007] A further technical solution is that the dynamic data table includes a data layer, a tagging layer, and a tracking layer. The data layer has several time cells arranged vertically, and these time cells are arranged in the order they were entered. Each time cell is associated with at least one data cell, which is used to enter case data, dynamic data, and manual data. The tagging layer includes a first associated cell and a risk tagging cell associated with the first associated cell. Each risk tagging cell is associated with a risk trigger cell, a duration cell, and a risk value cell. The first associated cell is associated with the time cells in the data layer. The tracking layer includes a second associated cell and a remaining time cell, a monitoring trigger cell, and a filter cell associated with the second associated cell. The second associated cell is associated with the first associated cell in the tagging layer. The remaining time cell collects the duration cell and its corresponding time cell to form a countdown. The monitoring trigger cell collects the risk trigger from the risk trigger cell corresponding to the duration and fills it into the monitoring trigger cell. If the time in the remaining time cell is not zero, the filter cell is filled with the number 1; if the time in the remaining time cell is zero, the filter cell is filled with the number 0.

[0008] A further technical solution involves a fusion processing module that organizes and merges case data, dynamic data, and manual data based on a timeline, and fills the corresponding positions in the dynamic data table. The steps include: Step S101, receiving case data, dynamic data, and manual data through a multi-source data acquisition module and performing preliminary verification; Step S102, assigning time tags to the case data, dynamic data, and manual data based on the time information on the data; and Step S103, collecting the time of the time tags and filling them into time cells, and filling case data, dynamic data, and manual data with the same time tag into multiple data cells associated with the same time cell.

[0009] A further technical solution involves the AI ​​analysis module analyzing the dynamic data table entered in step S103. This includes: Step S201, extracting multi-dimensional features related to the pregnant woman's physiological state, fetal condition, and potential risks. These features include vital sign trends, abnormal test results, symptom description keywords, past medical history correlation features, and equipment monitoring data fluctuation features; Step S202, based on a pre-trained obstetric risk identification AI model, analyzing the extracted multi-dimensional features, identifying and marking potential risk points, and writing the identified risk point types into the corresponding time-point associated risk marker cells; Step S203, for each risk point marked in step S202... Step S204: Analyze the data characteristics before or during the occurrence of the risk point, identify and mark the risk triggers that may trigger the risk point, and write the information related to the identified risk triggers into the risk trigger cell associated with the corresponding risk point; Step S205: Predict the possible duration or evolution window of the risk point through the risk model, calculate the time span from the start to the expected end of the risk point, and write the duration information into the duration cell associated with the corresponding risk point; Step S206: Calculate a risk value for each risk point based on factors such as the type, severity, clarity of the risk trigger, length of the risk duration, and current gestational age, and fill the risk value into the corresponding risk value cell.

[0010] A further technical solution is that the steps for the AI ​​early warning module to generate a preliminary early warning report include: step S301, real-time monitoring of the updates of data cells in the dynamic data table and extraction of new data features; step S302, matching the new data features with risk triggers associated with all filtered cell values ​​of 1 in the marker layer; step S303, if the new data features meet any risk trigger condition, the early warning mechanism is triggered and a preliminary early warning report is generated.

[0011] A further technical solution involves the following steps in which the review module receives a preliminary warning report and formulates a second manual review plan based on the risk level of the preliminary warning report: Step S401, parsing the risk value cell data in the preliminary warning report and dividing the risk values ​​into different risk levels, including low-risk, medium-risk, and high-risk ranges; Step S402, matching review time limits for different risk levels: low-risk reports are reviewed within 24 hours, medium-risk reports within 4 hours, and high-risk reports trigger an immediate response mechanism; Step S403, dynamically matching the reviewer's qualifications based on the risk type and level, requiring reviewers with high qualifications for low-risk reports. For general obstetrics attending physicians, intermediate-risk reports require dual review, with one physician having qualifications no lower than that of a general obstetrics attending physician and the other being a specialist attending physician matching the risk level. High-risk reports require multiple reviewers, including at least one associate chief obstetrics physician with qualifications no lower than that of an associate chief obstetrics physician and one associate chief physician matching the risk level. Step S404: Link to the historical case database, collect the handling records and expert comments of similar risk cases as a reference for review. Step S405: Generate a review task package, including the original text of the preliminary warning report, a matching review process list, and a historical case reference set, and assign it to the to-do queue of the corresponding qualified reviewers.

[0012] A further technical solution is that the preliminary verification of the received case data, dynamic data and manual data in step S101 includes checking whether all case data, dynamic data and manual data contain required fields, verifying whether key indicators are missing for dynamic data, and verifying whether there are empty values ​​or invalid inputs for manual data.

[0013] Compared with existing technologies, the beneficial effects of this invention are: 1. By integrating electronic medical records, wearable devices, and manually input data, a comprehensive health profile of pregnant women and new mothers is constructed, overcoming the limitations of a single data source. Based on timeline-based data processing and fusion, it can accurately capture the evolution trends and correlations of physiological parameters, symptoms, and signs, and identify subtle changes and early risk signals that are easily missed by traditional intermittent prenatal checkups. 2. The AI ​​analysis module can not only identify explicit risk points, but also delve deeper into potential risk triggers, quantify risk values, and determine the risk duration window, significantly improving the sensitivity of risk identification and achieving "early detection and early warning" of risks. 3. Based on the risk level and nature, the system intelligently generates preliminary precaution reports and preliminary warning reports. Preliminary precaution reports inform mothers of necessary precautions, enabling them to avoid potential risks in advance. Preliminary warning reports allow for rapid response and appropriate measures once a risk trigger is established. This condition-triggered mechanism makes warnings more targeted, real-time, and clinically valuable, avoiding unnecessary panic. Preliminary warning reports directly include the warning level and corresponding risk management plan, providing frontline medical staff with immediate and standardized decision support and shortening the time window from risk identification to action. 4. The review module intelligently matches different review plans based on report type and warning level. This ensures that reports of different risk levels receive the most appropriate review resources. The allocation module accurately assigns reports to the most suitable reviewers based on the complexity and professional requirements of the review plan, greatly improving the efficiency of human resource utilization. 5. This application utilizes AI to complete the initial risk identification, quantification, classification, and report generation, significantly reducing the burden of manual screening of massive amounts of data. Human review focuses on verifying, correcting, and making final decisions regarding the AI ​​results, applying human expertise to the most critical stages, forming a highly efficient collaborative model of "machine rapid screening, expert oversight." Doctors receive risk assessment reports based on comprehensive and dynamic data analysis, including risk points, triggers, duration, quantification level, and treatment plans, assisting them in making more scientific and individualized clinical decisions. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0015] Example 1:

[0016] An AI-based intelligent early warning system for obstetric risks includes: a data table creation module for creating dynamic data tables; a multi-source data acquisition module for collecting case data from electronic medical record systems, dynamic data from wearable medical devices, and manually entered data; a fusion processing module for organizing and fusing case data, dynamic data, and manual data based on a timeline and filling them into the corresponding positions in the dynamic data table; an AI analysis module for analyzing the dynamic data table, marking risk points, risk triggers, and the duration of risk points, and assigning risk values ​​to risk points; and an AI early warning module for generating preliminary precaution reports based on risk points, risk triggers, and the duration of risk points. If a risk trigger is triggered within the duration of a risk point, a preliminary early warning report is generated, including the warning level and risk assessment criteria. The system comprises a risk management scheme; an audit module for receiving preliminary precaution reports and preliminary warning reports, matching the preliminary precaution reports with a first manual audit scheme based on a preset audit scheme, and formulating a second manual audit scheme for preliminary warning reports based on risk levels; an allocation module for receiving preliminary precaution reports and their matching first and second manual audit schemes, and allocating them to corresponding auditors for manual review based on the first and second manual audit schemes; a manual audit module for receiving allocated preliminary precaution reports and preliminary warning reports, and reviewing and modifying them to form final precaution reports and final warning reports; and a receiving module for receiving final precaution reports and final warning reports. The system includes a user terminal, a medical staff terminal, and a server terminal. A multi-source data acquisition module is configured on the user terminal and the medical staff terminal to collect case data, dynamic data, and manual data. Manual data can be manually entered through either the user terminal or the medical staff terminal. The receiving module is configured on the user terminal, and the manual audit module is configured on the medical staff terminal. The data table creation module, fusion processing module, AI analysis module, AI warning module, audit module, and allocation module are configured on the server terminal.

[0017] The dynamic data table includes a data layer, a tagging layer, and a tracking layer. The data layer consists of several time cells arranged vertically, with each time cell linked to at least one data cell. These data cells are used to input case data, dynamic data, and manual data. The tagging layer includes a first associated cell and a risk tagging cell associated with it. Each risk tagging cell is associated with a risk trigger cell, a duration cell, and a risk value cell. The first associated cell is linked to the time cells in the data layer. The tracking layer includes a second associated cell and associated with it a remaining time cell, a monitoring trigger cell, and a filter cell. The second associated cell is linked to the first associated cell in the tagging layer. The remaining time cell collects the duration cell and its corresponding time cell to form a countdown. The monitoring trigger cell collects the risk trigger from the risk trigger cell corresponding to the duration and fills it into the monitoring trigger cell. If the time in the remaining time cell is not zero, the filter cell contains the number 1; if the time in the remaining time cell is zero, the filter cell contains the number 0. The Remaining Time cell calculates and displays the remaining monitoring time for the risk point based on the start time (from the time cell in the data layer) and the predicted duration (from the duration cell) of the associated risk marker cell; the Monitoring Trigger cell stores or references the content of the associated risk trigger cell; the Monitoring Status cell indicates whether the risk point is under monitoring: when the remaining time is greater than zero, the status is active (e.g., value 1), indicating that the corresponding trigger needs to be monitored; when the remaining time is zero, the status is closed (e.g., value 0).

[0018] The fusion processing module organizes and merges case data, dynamic data, and manual data based on a timeline, and fills them into the corresponding positions in the dynamic data table. The steps include: Step S101, receiving case data, dynamic data, and manual data through the multi-source data acquisition module and performing preliminary verification; Step S102, assigning time tags to the case data, dynamic data, and manual data based on the time information on them; Step S103, collecting the time of the time tags and filling them into the time cells, and filling case data, dynamic data, and manual data with the same time tag into multiple data cells associated with the same time cell. For the same time tag, there may be one or more data points; therefore, flexibly increasing the number of data cells associated with the time cell according to the amount of data ensures that the time consistency of various types of data within these data cells is maintained.

[0019] The AI ​​analysis module analyzes the dynamic data table entered in step S103, including the following steps: Step S201, extracting multi-dimensional features related to the pregnant woman's physiological state, fetal condition, and potential risks. These multi-dimensional features include vital sign trends, abnormal test indicators, symptom description keywords, past medical history correlation features, and equipment monitoring data fluctuation features; Step S202, based on a pre-trained obstetric risk identification AI model, analyzing the extracted multi-dimensional features, identifying and marking potential risk points, and writing the identified risk point types into the risk marker cells associated with the corresponding time points; Step S203, analyzing the occurrence of each risk point marked in step S202. The data characteristics before or at the time of occurrence are used to identify and mark the risk triggers that may cause the risk point, and the information related to the identified risk triggers is written into the risk trigger cell associated with the corresponding risk point; Step S204, the possible duration or evolution window of the risk point is predicted by the risk model, the time span from the start to the expected end of the risk point is calculated, and the duration information is written into the duration cell associated with the corresponding risk point; Step S205, based on factors such as the type, severity, clarity of the risk trigger, length of the risk duration, and current gestational age, a risk value is calculated for each risk point, and the risk value is filled into the corresponding risk value cell. When performing multi-dimensional feature extraction, a pre-trained medical BERT model can be used. The type of pre-trained obstetric risk identification AI model can be one of decision trees, random forests, SVMs, or deep learning models, and the training data source can be anonymized historical medical records. The risk value can be assigned a fixed risk value to each risk point by an experienced expert group, or each risk point can be assigned a score for each factor during the calculation process, and each factor can be assigned a weight, and the final risk value is obtained through a weighted calculation.

[0020] The steps for the AI ​​early warning module to generate a preliminary early warning report include: Step S301, real-time monitoring of data cell updates in the dynamic data table and extraction of new data features; Step S302, matching the new data features with risk triggers associated with all filtered cell values ​​of 1 in the marker layer; Step S303, if the new data features meet any risk trigger condition, triggering the early warning mechanism and generating a preliminary early warning report.

[0021] The review module receives the preliminary warning report and formulates a second manual review plan based on the risk level of the preliminary warning report. The steps include: Step S401, parsing the risk value cell data in the preliminary warning report and dividing the risk values ​​into different risk levels, including low, medium, and high risk; Step S402, matching review time limits for different risk levels: low-risk reports are reviewed within 24 hours, medium-risk reports within 4 hours, and high-risk reports trigger an immediate response mechanism; Step S403, dynamically matching the reviewer's qualifications based on the risk type and level, requiring the reviewer's qualifications for low-risk reports to be no lower than those of a general obstetrics specialist. For intermediate-risk reports, dual review is required. One physician must be at least a qualified attending obstetrician, and the other must be a specialist attending physician matching the risk level. For high-risk reports, multiple review is required, including at least one associate chief obstetrician and one associate chief physician matching the risk level. Step S404: Link the historical case database to collect the handling records and expert comments of similar risk cases as a reference for review. Step S405: Generate a review task package, which includes the original preliminary warning report, a matching review process list, and a historical case reference set, and assign it to the to-do queue of the corresponding qualified reviewers.

[0022] Step S101 involves preliminary verification of the received case data, dynamic data, and manual data, including checking whether all case data, dynamic data, and manual data contain required fields, verifying whether key indicators are missing from the dynamic data, and verifying whether there are any empty values ​​or invalid inputs in the manual data.

[0023] Example 2: System Example.

[0024] Hardware environment:

[0025] Deployed on the hospital's private cloud server (16 CPU cores / 64GB memory / 1TB storage).

[0026] Access devices: hospital electronic medical record system (EMR), wearable devices for pregnant women (fetal heart monitor / blood pressure bracelet), and mobile data entry terminal.

[0027] Example of dynamic data table construction: Patient information: 32 weeks + 2 days pregnant, primiparous, BMI 28.

[0028] Table 1: Data Table Structure (Taking Pregnancy-Induced Hypertension Risk Monitoring as an Example)

[0029] Demonstration of the entire data processing workflow.

[0030] 1. Multi-source data acquisition and fusion.

[0031] Electronic medical record data: pregnancy history (pre-pregnancy hypertension), laboratory data (urine protein +).

[0032] Wearable devices: Real-time blood pressure (145 / 95→150 / 100mmHg), fetal heart rate (160bpm).

[0033] Manual entry: Patient's chief complaint: "Blurred vision".

[0034] Merge operation: Integrate into the dynamic data table by timestamp, triggering step S101 verification (required field integrity verification).

[0035] 2. AI analysis module execution (steps S201-S205).

[0036] Feature extraction: Blood pressure increased by 30% within 24 hours, proteinuria, and neurological symptom cluster.

[0037] Risk marker: Mark "high risk of preeclampsia" at 09:00.

[0038] Trigger correlation: Sudden rise in blood pressure + new symptoms → Enter into the risk trigger cell.

[0039] Duration prediction: 72-hour high-risk window calculated based on clinical guideline model.

[0040] Risk value calculation: Risk = 0.4 × (blood pressure score) + 0.3 × (symptom score) + 0.2 × (test score) + 0.1 × (gestational week score) = 82.

[0041] 3. AI warning triggered (steps S301-S303).

[0042] 10:00 New data: Blood pressure 158 / 102 mmHg + worsening blurred vision.

[0043] Match the monitoring trigger cell conditions → Generate a preliminary early warning report.

[0044] High-risk warning (Level I) **Risk type**: Acute preeclampsia **Treatment plan**: 1. Immediate bed rest + intravenous magnesium sulfate infusion 2. Monitor blood pressure every 15 minutes 3. Prepare an emergency cesarean section plan.

[0045] 4. Tiered review mechanism (steps S401-S405).

[0046] Risk value 82 → High risk level.

[0047] Audit Plan:

[0048] Review team: Associate Chief Physician of Obstetrics and Associate Chief Physician of Neurology.

[0049] Timeframe: Immediate response (<10 minutes).

[0050] Historical case reference: A case from 2023 (92% similarity).

[0051] 5. Output is manually reviewed.

[0052] The doctor added: Additional fundus examination and adjustment of magnesium sulfate dosage.

[0053] The final warning report adds:

[0054] "Fundus examination showed retinal artery spasm, and it is recommended that the assessment window for terminating the pregnancy be shortened to 24 hours."

[0055] Although the invention has been described herein with reference to several illustrative embodiments, it should be understood that many other modifications and implementations can be devised by those skilled in the art, which will fall within the scope and spirit of the principles disclosed herein. More specifically, various variations and modifications can be made to the components and / or layout of the subject matter combination within the scope of the disclosure and claims. Besides variations and modifications to the components and / or layout, other uses will be apparent to those skilled in the art.

Claims

1. An AI-based intelligent early warning system for obstetric risks, characterized in that, include: The data table creation module is used to create dynamic data tables; The multi-source data acquisition module is used to collect case data from the electronic medical record system, dynamic data from wearable medical devices, and manually entered data. The fusion processing module organizes and merges case data, dynamic data, and manual data based on the timeline, and fills them into the corresponding positions in the dynamic data table; the AI ​​analysis module is used to analyze the dynamic data table, mark risk points, risk causes, and the duration of risk points, and assign risk values ​​to risk points. The AI ​​early warning module generates a preliminary precaution report based on the risk point, risk trigger, and duration of the risk point. If the risk trigger is triggered within the duration of the risk point, a preliminary early warning report is generated, which includes the early warning level and risk handling plan. The review module is used to receive preliminary notice reports and preliminary warning reports. The preliminary notice reports are matched with the first manual review plan according to the preset review plan, and the preliminary warning reports are formulated with the second manual review plan according to the risk level. The allocation module is used to receive the preliminary notice report and the matching first manual review plan, as well as the preliminary warning report and the matching second manual review plan, and to allocate the report to the corresponding reviewer for manual review according to the first and second manual review plans. The manual review module is used to receive the assigned preliminary precaution reports and preliminary warning reports, and to review and modify them to form the final precaution reports and final warning reports. The receiving module is used to receive the final precautions report and the final warning report.

2. The AI-based intelligent early warning system for obstetric risks according to claim 1, characterized in that: The dynamic data table includes a data layer, a tagging layer, and a tracking layer. The data layer consists of several time cells arranged vertically, with each time cell linked to at least one data cell. These data cells are used to input case data, dynamic data, and manual data. The tagging layer includes a first associated cell and risk tagging cells associated with it. Each risk tagging cell is associated with a risk cause cell, a duration cell, and a risk value cell. The first associated cell is linked to the time cells in the data layer. The tracking layer includes a second associated cell, as well as a remaining time cell, a monitoring trigger cell, and a filter cell associated with the second associated cell. The second associated cell is associated with the first associated cell of the tagging layer. The remaining time cell collects the duration cell and its corresponding time cell to form a time countdown. The monitoring trigger cell collects the risk triggers in the risk trigger cells corresponding to the duration and fills them into the monitoring trigger cell. If the time in the remaining time cell has not reached zero, the filter cell is filled with the number 1. If the time in the remaining time cell has reached zero, the filter cell is filled with the number 0.

3. The AI-based intelligent early warning system for obstetric risks according to claim 2, characterized in that: The fusion processing module organizes and merges case data, dynamic data, and manual data based on the timeline, and fills them into the corresponding positions in the dynamic data table. The steps include: Step S101, receiving case data, dynamic data, and manual data through the multi-source data acquisition module and performing preliminary verification; Step S102, assigning time tags to case data, dynamic data, and manual data based on the time information on the case data, dynamic data, and manual data; Step S103, collecting the time of the time tags and filling them into the time cells, and filling case data, dynamic data, and manual data with the same time tag into multiple data cells associated with the same time cell.

4. The AI-based intelligent early warning system for obstetric risks according to claim 3, characterized in that: The AI ​​analysis module analyzes the dynamic data table after it is filled in step S103. The steps include step S201, extracting multi-dimensional features related to the physiological state of pregnant women, fetal condition and potential risks. The multi-dimensional features include vital sign trends, abnormal test indicators, symptom description keywords, past medical history related features and equipment monitoring data fluctuation features. Step S202: Based on the pre-trained obstetric risk identification AI model, analyze the extracted multi-dimensional features, identify and mark potential risk points, and write the identified risk point types into the risk mark cells associated with the corresponding time points; Step S203: For each risk point marked in Step S202, analyze its data characteristics before or when it occurs, identify and mark the risk triggers that will trigger the risk point, and write the information related to the identified risk triggers into the risk trigger cells associated with the corresponding risk point; Step S204: For each risk point, predict its possible duration or evolution window using a risk model, calculate the time span from the start to the expected end of the risk point, and write the duration information into the duration cell associated with the corresponding risk point; Step S205: Calculate a risk value for each risk point based on its type, severity, the clarity of the risk trigger, the length of the risk duration, and the current gestational age, and fill the risk value into the corresponding risk value cell.

5. The AI-based intelligent early warning system for obstetric risks according to claim 3, characterized in that: The steps for the AI ​​early warning module to generate a preliminary early warning report include: Step S301, real-time monitoring of data cell updates in the dynamic data table and extraction of new data features; Step S302, matching the new data features with risk triggers associated with all filtered cell values ​​of 1 in the marker layer; Step S303, if the new data features meet any risk trigger condition, triggering the early warning mechanism and generating a preliminary early warning report.

6. The AI-based intelligent early warning system for obstetric risks according to claim 1, characterized in that: The review module receives the preliminary warning report and formulates a second manual review plan based on the risk level of the preliminary warning report. The steps include: Step S401, parsing the risk value cell data in the preliminary warning report and dividing the risk values ​​into different risk levels, including low, medium, and high risk; Step S402, matching review time limits for different risk levels: low-risk reports are reviewed within 24 hours, medium-risk reports within 4 hours, and high-risk reports trigger an immediate response mechanism; Step S403, dynamically matching the reviewer's qualifications based on the risk type and level, requiring the reviewer's qualifications for low-risk reports to be no lower than those of a general obstetrics specialist. For intermediate-risk reports, dual review is required. One physician must be at least a qualified attending obstetrician, and the other must be a specialist attending physician matching the risk level. For high-risk reports, multiple review is required, including at least one associate chief obstetrician and one associate chief physician matching the risk level. Step S404: Link the historical case database to collect the handling records and expert comments of similar risk cases as a reference for review. Step S405: Generate a review task package, which includes the original preliminary warning report, a matching review process list, and a historical case reference set, and assign it to the to-do queue of the corresponding qualified reviewers.

7. The AI-based intelligent early warning system for obstetric risks according to claim 3, characterized in that: Step S101 involves preliminary verification of the received case data, dynamic data, and manual data, including checking whether all case data, dynamic data, and manual data contain required fields, verifying whether key indicators are missing from the dynamic data, and verifying whether there are any empty values ​​or invalid inputs in the manual data.

Images