A fetal risk monitoring method and system based on gestational age physiological characteristics

Abstract

A method for fetal risk monitoring based on gestational physiological characteristics includes the following steps: collecting physiological indicator data of pregnant women during pregnancy; preprocessing the physiological indicator data; establishing a multimodal model, inputting the preprocessed physiological indicator data into the multimodal model, which outputs the types of abnormal physiological indicators, as well as the fetal risk score and risk level; implementing targeted non-invasive intervention measures based on the risk level and the types of abnormal physiological indicators; continuously monitoring the physiological indicator data after intervention and dynamically updating the risk assessment results. The fetal risk monitoring method and system based on gestational physiological characteristics provided by this invention can monitor the health risks of the fetus based on the physiological characteristics of pregnant women during pregnancy, thereby facilitating prediction before health risks to the fetus occur and helping medical staff gain sufficient intervention time.

Description

Technical Field

[0001] This invention relates to the field of fetal risk monitoring technology, specifically to a method and system for fetal risk monitoring based on gestational age physiological characteristics. Background Technology

[0002] Fetal risk monitoring is a core component of prenatal care. Its main purpose is to promptly detect various abnormal risks during fetal development (such as premature birth, intrauterine growth retardation, placental dysplasia, fetal hypoxia, and fetal malabsorption), providing accurate evidence for clinical intervention, reducing the incidence of adverse pregnancy outcomes, and ensuring maternal and infant safety. Currently, commonly used fetal risk monitoring methods are mainly divided into two categories: one is invasive monitoring, represented by amniocentesis, which involves extracting amniotic fluid samples from the pregnant woman to detect fetal chromosomal abnormalities. Although this method has high accuracy in detecting chromosomal abnormalities, it carries a certain risk of miscarriage and must be performed by professional medical staff in a hospital, posing a certain risk of physical trauma to the pregnant woman. Another type is non-invasive monitoring, including Down syndrome screening, non-invasive prenatal testing (NIPT), ultrasound examination, and fetal heart rate monitoring. Down syndrome screening and NIPT are mainly used to screen for fetal chromosomal abnormalities (such as trisomy 21 and trisomy 18), but they have a certain false positive and false negative rate, and the testing cycle is relatively long, making real-time monitoring impossible. Ultrasound examination is mainly used to screen for fetal structural malformations and monitor fetal growth indicators (such as biparietal diameter and femur length), but it can only be detected after obvious signs of fetal developmental abnormalities appear, and cannot predict risks in advance.

[0003] With the rapid development of home medical devices, some non-invasive physiological indicator collection devices have begun to be applied to pregnancy monitoring, such as home fetal heart rate monitors and blood pressure monitors. However, these devices are mostly limited to single indicator monitoring and have not formed a systematic risk assessment and intervention system. They can only provide basic indicator data and cannot achieve correlation analysis between indicators and fetal risk, let alone provide targeted intervention measures for abnormal indicators. At the same time, existing fetal risk monitoring technologies generally follow the traditional approach of "directly monitoring the fetus," ignoring the two-way physiological interaction between the fetus and the mother. During fetal development, the fetus secretes specific factors through the placenta, regulating the pregnant woman's hormone levels, gut microbiota balance, autonomic nervous system function, and gait characteristics. These atypical physiological changes in the pregnant woman are all closely and implicitly related to the fetal development status and risk. Therefore, there is a need for a method that can monitor fetal risk based on the pregnant woman's physiological characteristics during pregnancy. This invention addresses this technical problem. Summary of the Invention

[0004] This invention provides a method and system for fetal risk monitoring based on the physiological characteristics of gestational weeks. It can monitor the health risks of the fetus according to the physiological characteristics of the pregnant woman during the gestational period, thereby facilitating the prediction of fetal health risks before they occur and helping medical staff gain sufficient time for intervention.

[0005] A method for fetal risk monitoring based on gestational age physiological characteristics includes the following steps:

[0006] Collect physiological data of pregnant women during pregnancy;

[0007] The physiological index data are preprocessed;

[0008] A multimodal model is established, and the preprocessed physiological index data is input into the multimodal model. The multimodal model is used to output the abnormality type of physiological index, as well as the risk score and risk level of the fetus.

[0009] Based on the risk level and the type of abnormality in the physiological indicators, implement targeted non-invasive intervention measures;

[0010] Continuously monitor physiological indicators after intervention and dynamically update risk assessment results.

[0011] Furthermore, the physiological indicators include salivary hormone indicators, gut microbiota indicators, gait parameters, and skin conductance indicators. These physiological indicators are collected specifically according to different stages of gestation, and these different stages of gestation correspond to key stages of fetal development.

[0012] Furthermore, the physiological indicator data are collected using a non-invasive home-use collection device, which includes a saliva collector, an intestinal flora collection box, an intelligent gait collection device, and a skin conductance collection patch. The collection device is used to collect data in real time and synchronize the data to a data processing terminal.

[0013] Furthermore, the data preprocessing includes removing outliers from physiological indicator data, eliminating short-term data fluctuations through time-series smoothing, normalizing the obtained data, and finally aligning the processed data with gestational age.

[0014] Furthermore, the multimodal model includes a temporal model and an association model. The temporal model is used to capture the temporal variation characteristics of physiological indicators and output a temporal risk score. The association model is used to mine the implicit associations between physiological indicators and output an association risk score. The temporal risk score and the association risk score are fused through an attention mechanism to obtain the abnormality type of physiological indicators and the final fetal risk score.

[0015] Furthermore, the attention mechanism fusion adopts a preset weight allocation method, the preset weights are determined through clinical data calibration, and the sum of the weights of the temporal risk score and the associated risk score is 1.

[0016] Furthermore, the fetal risk score is combined with a preset threshold, and the fetal risk score is divided into multiple levels to obtain the fetal risk level. Different fetal risk levels correspond to different monitoring methods and intervention strategies.

[0017] Furthermore, the targeted non-invasive intervention measures correspond one-to-one with the types of abnormal physiological indicators.

[0018] A fetal risk monitoring system based on gestational age physiological characteristics, based on the aforementioned fetal risk monitoring method based on gestational age physiological characteristics, includes a data acquisition module, a data preprocessing module, a risk prediction module, an intervention guidance module, and a closed-loop monitoring module.

[0019] Furthermore, it also includes a monitoring equipment kit and a mobile terminal APP. The mobile terminal APP is used to display monitoring data, risk scores and intervention suggestions in real time, and to connect with medical diagnosis and treatment systems to achieve data linkage.

[0020] The technical effects of this invention are as follows:

[0021] This invention uses non-invasive equipment to collect physiological characteristic data of pregnant women, avoiding the miscarriage risks associated with invasive methods such as amniocentesis, thus helping to ensure the safety of both the pregnant woman and the fetus. Simultaneously, this method collects various physiological characteristic data throughout gestational age and analyzes this data using a multimodal model that integrates time-series and correlation models. This not only captures the temporal changes in physiological indicators but also uncovers implicit correlations between different physiological indicators, facilitating timely prediction and detection of fetal health risks and achieving comprehensive monitoring throughout pregnancy. Attached Figure Description

[0022] Figure 1 This is a flowchart of the process of the present invention. Detailed Implementation

[0023] The technical solution of the present invention will be clearly and completely described below with reference to specific embodiments and accompanying drawings.

[0024] See Figure 1 A method for fetal risk monitoring based on gestational age physiological characteristics includes the following steps:

[0025] Collect physiological data of pregnant women during pregnancy;

[0026] Preprocess the physiological indicator data;

[0027] A multimodal model is established. The preprocessed physiological index data is input into the multimodal model, which is used to output the abnormality type of physiological index, as well as the risk score and risk level of the fetus.

[0028] Based on the risk level and the type of abnormal physiological indicators, implement targeted non-invasive intervention measures;

[0029] Continuously monitor physiological indicators after intervention and dynamically update risk assessment results.

[0030] Furthermore, physiological indicators include salivary hormone levels, gut microbiota levels, gait parameters, and skin conductance. These physiological indicators are collected specifically according to different stages of gestation, with each stage corresponding to a critical stage of fetal development.

[0031] In this embodiment, the different stages of gestational age are divided as follows:

[0032] The first stage (placental formation period): This corresponds to the stage where the placenta begins to form and gradually matures. During this stage, the placental development directly affects the oxygen and nutrient supply to the fetus. Atypical physiological indicators collected include salivary hormone-related indicators and gut microbiota-related indicators. Salivary hormone-related indicators include the concentrations and fluctuations of salivary cortisol and salivary progesterone. During placental formation, the fetus secretes specific factors through the placenta to regulate the pregnant woman's progesterone and cortisol levels. The fluctuations in these concentrations are positively correlated with the placental vascular development status; abnormal fluctuations indicate poor placental development, which may lead to fetal hypoxia. Gut microbiota-related indicators include the abundance of Bifidobacteria, Firmicutes, and Bacteroidetes, and the Firmicutes / Bacteroidetes ratio (F / B ratio). Gut microbiota balance affects the pregnant woman's metabolic state and inflammation level, thereby affecting placental vascular development. Microbiota imbalance increases the risk of fetal malabsorption.

[0033] The second stage (rapid fetal development period): This stage corresponds to the rapid development of fetal weight and organs. During this stage, the pregnant woman's physical condition (such as center of gravity and autonomic nervous system function) will change significantly due to fetal development and is closely related to the fetal development status. Atypical physiological indicators collected include gait parameters and skin conductance parameters. Gait parameters include stride length, gait frequency, and center of gravity shift. Rapid fetal development leads to increased maternal weight and a forward shift of the center of gravity. Abnormal changes in gait parameters reflect a decrease in the pregnant woman's core strength, which can lead to uterine compression of the fetus and increase the risk of intrauterine growth retardation. Skin conductance parameters include skin conductivity and response latency. Skin conductance is related to the pregnant woman's autonomic nervous system function. Abnormally elevated levels indicate a stress response in the pregnant woman, which can lead to abnormal fetal heart rate and increase the risk of premature birth.

[0034] The third stage (fetal engagement / labor preparation period): This stage corresponds to the gradual engagement of the fetus into the pelvis, preparing for labor. During this stage, the pregnant woman's physiological state changes more significantly and is closely related to risks such as preterm birth and intrauterine distress. Atypical physiological indicators collected include diurnal differences in skin conductance, dynamic changes in gait parameters, and the diurnal rhythm of salivary hormones. Abnormal diurnal differences in skin conductance and the diurnal rhythm of salivary cortisol indicate elevated stress levels in the pregnant woman, which are important warning signs of preterm birth; abnormal gait center of gravity shift rate indicates abnormal fetal engagement.

[0035] Furthermore, physiological data are collected using non-invasive home-use collection devices, including saliva collectors, gut microbiota collection boxes, intelligent gait collection devices, and skin conductance collection patches. These devices are used to collect data in real time and synchronize the data to a data processing terminal.

[0036] The data acquisition equipment and methods used in this embodiment are as follows:

[0037] Saliva Collection Device: This device uses a saliva collector with a built-in detection module to quickly detect the concentrations of cortisol and progesterone in saliva. Collection is performed every morning on an empty stomach (without food or water), collecting a preset volume of saliva sample. The built-in detection module completes the test within a preset time. The data is synchronized in real-time to a mobile app via Bluetooth, allowing pregnant women to view and track the data. This collection device is compact, easy to operate, requires no specialized knowledge, is reusable or disposable, commercially available, and suitable for pregnant women at all gestational ages.

[0038] Intestinal flora collection equipment: This device uses an intestinal flora collection kit, which includes specialized sampling tools, microbial extraction reagents, and sample preservation devices. Collection is conducted at a fixed time each week, collecting a predetermined amount of fecal samples, placing them in the extraction reagents within the collection kit, shaking well, and then sending them to a professional testing center via a pre-designated courier service. The testing center completes the analysis of flora abundance and the F / B ratio within a predetermined timeframe, and the results are simultaneously sent to a mobile app. This collection kit effectively avoids sample contamination, and the testing service has been adopted by numerous hospitals nationwide. The test results can be used as a clinical reference.

[0039] Intelligent gait data collection device: Utilizing smart insoles or smart bracelets, this device incorporates pressure and acceleration sensors to collect real-time gait parameters such as stride length, cadence, and center of gravity shift during a pregnant woman's walk. Data collection occurs at fixed times daily, with the pregnant woman wearing the device walking a preset number of steps at a constant speed. The device automatically filters out invalid data such as unstable walking, pauses, and turns. After collection, the data is automatically synchronized to a mobile app, which can perform preliminary analysis of the gait parameters and mark abnormal data. This device is suitable for everyday wear by pregnant women, does not interfere with normal activities, and offers high accuracy, accurately reflecting the dynamic changes in a pregnant woman's gait.

[0040] Skin conductance data acquisition device: Utilizing a medical-grade skin conductance patch, this device can be applied to the inside of a pregnant woman's wrist or other skin areas. It features a built-in acquisition module that can collect real-time data on skin conductivity, response latency, and other indicators. Data is collected by wearing the device at fixed times each day for a preset duration. The device automatically filters out abnormal data caused by factors such as sleep and exercise. After collection, the data is synchronized to a mobile app. The patch adheres firmly without damaging the skin and has a long battery life, meeting the needs of extended daily data collection.

[0041] Furthermore, data preprocessing includes removing outliers from physiological indicator data, eliminating short-term data fluctuations through time-series smoothing, normalizing the obtained data, and finally aligning the processed data with gestational age.

[0042] The specific data preprocessing process in this embodiment is as follows:

[0043] Outlier Removal: Statistical outlier handling methods (preferably IQR) are employed to screen all collected raw physiological indicator data, removing extreme outliers caused by equipment malfunctions, operational errors, temporary emotional fluctuations, etc. The specific process is as follows: First, the first quartile (Q1), third quartile (Q3), and interquartile range (IQR = Q3 - Q1) of each indicator are calculated. Then, outlier judgment criteria are set, removing outliers exceeding "Q1 - preset multiple × IQR" or "Q3 + preset multiple × IQR". For the removed outliers, linear interpolation is used for repair. This involves calculating the repaired value based on the valid data before and after the outlier, ensuring data continuity and integrity, and preventing outliers from interfering with subsequent modeling. The preset multiple is determined through large-scale clinical data calibration to ensure the accuracy of outlier judgment, neither omitting genuine outlier data nor misjudging normal physiological fluctuations as abnormal.

[0044] Time-series smoothing: The moving average method is used to smooth the data after outlier removal and repair, eliminating short-term fluctuations (such as temporary emotional fluctuations in pregnant women or abnormal indicators caused by short-term activities) while preserving long-term trends related to fetal development. The specific process is as follows: A preset window size is set, and the average of all valid data within the window is calculated for each data point as the smoothed data for that point. The preset window size is determined through clinical data validation, preferably one week (7 days). This window size effectively eliminates short-term fluctuations without losing the key trends of indicator changes with gestational age, ensuring that the smoothed data accurately reflects the long-term changes in the pregnant woman's physiological indicators and corresponds to the long-term trend of fetal development.

[0045] Data standardization: A normalization method is used to convert physiological indicator data of different dimensions (such as salivary hormone concentration in μg / dL, gait length in cm, and skin conductivity in μS) into standardized data within the same numerical range, eliminating dimensional differences and facilitating multi-indicator fusion modeling. The specific process is as follows: First, the normal range of each indicator within the corresponding preset gestational week segment is determined, obtaining the minimum value (x_min) and maximum value (x_max) of the indicator. Then, using a normalization formula, each raw data point is converted into standardized data within a preset range (preferably [0,1]). The normal range (x_min, x_max) of each indicator is determined through calibration using large-scale clinical data from normal pregnant women to ensure the scientific validity of the standardized data analysis without altering the actual clinical significance of the indicators.

[0046] Gestational age synchronization alignment: All preprocessed physiological indicator data are time-series aligned according to clinically standard gestational age (day 1 of the last menstrual period), forming a "gestational age-multi-indicator" time-series dataset. The specific processing procedure is as follows: Using the first day of the pregnant woman's last menstrual period as day 1, the gestational age corresponding to each collected data point is calculated. All indicator data for the same gestational age are correlated to form a complete time-series dataset. If an indicator is not collected on a certain day (e.g., the pregnant woman forgets to collect a saliva sample), linear interpolation is used to supplement the data, ensuring the completeness and consistency of the time-series dataset and providing reliable support for subsequent AI models to capture the time-series changes of indicators. Simultaneously, the aligned time-series data is checked against the pregnant woman's clinical prenatal examination records to ensure that the gestational age matches the actual gestational age, avoiding risk assessment errors caused by gestational age calculation deviations.

[0047] Furthermore, the multimodal model includes a temporal model and an association model. The temporal model is used to capture the temporal variation characteristics of physiological indicators and output a temporal risk score, while the association model is used to mine the implicit associations between physiological indicators and output an association risk score. The temporal risk score and the association risk score are fused through an attention mechanism to obtain the abnormality type of physiological indicators and the final fetal risk score.

[0048] The analysis method for the multimodal model in this embodiment is as follows:

[0049] Temporal Model: The Transformer temporal model is preferred, employing an "input layer → positional encoding → multi-head self-attention layer → FeedForward layer → output layer" structure. Its core function is to process "gestational week-multiple indicators" temporal datasets, capturing the changing trends of various physiological indicators with gestational week and outputting a temporal risk score. Positional encoding marks the gestational week position of each data point, ensuring the model accurately identifies the temporal variation patterns of the indicators. The multi-head self-attention layer mines the temporal correlations between different gestational weeks and different indicators, quantifying the correlation strength of data from different gestational weeks and capturing the dynamic changes of indicators with gestational week. The FeedForward layer performs non-linear transformations on the features output by the attention layer, improving the model's fitting ability. The output layer outputs the temporal risk score, which reflects the contribution of temporal changes in physiological indicators to fetal risk. All model parameters (such as learning rate, number of iterations, number of attention heads, etc.) are determined through large-scale clinical data calibration to ensure the model accurately captures the correlation between temporal changes in indicators and fetal risk.

[0050] Association Model: A GNN (Graph Neural Network) association model is preferred, employing a structure of "node embedding layer → graph convolutional layer → pooling layer → output layer". Its core function is to construct an association network between physiological indicators, treating each indicator as a node, uncovering implicit interactions between indicators, and outputting an association risk score. Specifically, the node embedding layer converts each physiological indicator (node) into a computable feature vector for subsequent association analysis; the graph convolutional layer captures the association strength between nodes, quantifying the degree of association between different indicators using an adjacency matrix. The elements of the adjacency matrix are calculated using statistical correlation coefficients to reflect the correlation between different indicators; the pooling layer aggregates the features output by the graph convolutional layer to extract key association features; and the output layer outputs the association risk score, which reflects the contribution of abnormal associations between indicators to fetal risk. All model parameters (such as node embedding dimension, learning rate, and number of iterations) are calibrated using large-scale clinical data to ensure the model accurately captures the association between abnormal associations between indicators and fetal risk.

[0051] Attention mechanism fusion: A pre-defined weighting method is used to fuse the temporal risk score output by the temporal model and the correlation risk score output by the correlation model to obtain the final fetal risk score. The pre-defined weights are determined through large-scale clinical data calibration, assigning weights based on the contribution of the two models to fetal risk prediction. The sum of the weights of the temporal risk score and the correlation risk score is 1, ensuring the rationality and scientific validity of the fusion result. The final fused risk score integrates the dual risk indications of temporal changes in physiological indicators and abnormal correlations between indicators, providing a more comprehensive and accurate reflection of the fetus's actual risk status.

[0052] Furthermore, the attention mechanism fusion adopts a preset weight allocation method, the preset weights are determined through clinical data calibration, and the sum of the weights of the temporal risk score and the associated risk score is 1.

[0053] Furthermore, by combining the fetal risk score with a preset threshold, the fetal risk score is divided into multiple levels to obtain the fetal risk level. Different fetal risk levels correspond to different monitoring methods and intervention strategies.

[0054] Fetal risk levels are categorized into multiple grades based on the final risk score and preset thresholds, with a preferred classification of low, medium, and high risk. Different risk levels correspond to different monitoring and intervention strategies to ensure targeted and effective risk management. The preset thresholds are determined through ROC curve calibration using large-scale clinical data; the area under the ROC curve must meet the preset standard to ensure the accuracy of the threshold classification and avoid subjective judgment. The specific grade classifications and corresponding strategies are as follows:

[0055] Low risk: The final risk score is below the first preset threshold, indicating that there is no significant risk to the fetus. No targeted intervention is required; only monitoring at the routine frequency is necessary to regularly check changes in indicators and ensure normal fetal development. The preset threshold for this level is calibrated through clinical data to ensure that pregnant women in this level, when confirmed by traditional clinical gold standards (such as ultrasound and amniocentesis), achieve the preset standard in terms of accuracy in identifying no fetal risk, thus avoiding false negatives.

[0056] Medium risk: The final risk score falls between the first and second preset thresholds, indicating a potential risk to the fetus. Targeted non-invasive interventions are required, along with increased monitoring frequency, daily collection of physiological data, and real-time monitoring of changes in indicators and risk scores. The preset thresholds for this level are calibrated using clinical data to ensure that pregnant women in this category, confirmed by traditional clinical gold standards, have a high accuracy rate of potential risk, and that this type of risk can be reversed through non-invasive intervention.

[0057] High Risk: A final risk score exceeding the second pre-set threshold indicates a clear risk to the fetus, requiring immediate medical attention. Further diagnosis using traditional clinical gold standards (such as ultrasound, amniocentesis, and fetal heart rate monitoring) is necessary, along with clinical interventions to prevent further risk escalation. The pre-set threshold for this level is calibrated using clinical data to ensure that the accuracy rate of confirmation of a clear risk using traditional clinical gold standards for pregnant women in this level meets the pre-set standard, avoiding false positives and providing a reliable basis for emergency clinical intervention.

[0058] Furthermore, targeted non-invasive interventions correspond one-to-one with the types of abnormal physiological indicators.

[0059] The intervention measures correspond one-to-one with the types of abnormal physiological indicators. All are non-invasive methods commonly used in clinical practice, easy for pregnant women to operate, and readily available at home. No professional medical knowledge is required; pregnant women can perform them at home. All intervention measures have been validated by clinical data, achieving the preset effectiveness rate and reversing potential risks. Details are as follows:

[0060] Abnormal gut microbiota-related indicators: When the abundance of gut microbiota is abnormal or the F / B ratio is imbalanced, intervention measures include supplementing with probiotics and prebiotics. Among them, probiotics containing Bifidobacterium and Lactobacillus should be selected for pregnant women, and prebiotics such as inulin should be selected to promote the proliferation of probiotics. The dosage and frequency of administration are determined through clinical data calibration. This can effectively regulate the balance of gut microbiota, reduce metabolic inflammation in pregnant women, improve placental vascular development, and thus reverse the potential risk of fetal malabsorption.

[0061] Abnormal salivary hormone levels: When salivary cortisol levels are abnormally high or progesterone levels fluctuate abnormally, interventions include mindfulness meditation, listening to soothing music, and mood monitoring. Mindfulness meditation can utilize prenatal meditation courses built into mobile apps, with daily preset durations, which can effectively reduce stress responses and cortisol secretion in pregnant women. Mood monitoring can be achieved through devices such as smart bracelets, providing real-time reminders for pregnant women to regulate their emotions and avoid hormonal abnormalities caused by emotional fluctuations, thereby reducing the risk of fetal heart rate abnormalities and premature birth. When progesterone levels fluctuate abnormally, hormone levels can also be regulated and placental development improved through dietary adjustments and vitamin supplementation.

[0062] Abnormal gait parameters: When gait parameters such as center of gravity shift are abnormal, intervention measures include prenatal rehabilitation training (such as prenatal yoga) and the use of assistive devices (such as maternity support belts). Prenatal yoga can utilize core strength training courses for pregnant women built into mobile apps. Adhering to the preset duration daily can effectively improve the pregnant woman's core strength, adjust the center of gravity distribution, and reduce the pressure of the uterus on the fetus. Maternity support belts can share the weight of the uterus, adjust the pregnant woman's body posture, improve gait abnormalities, and thus reverse the potential risk of intrauterine growth retardation in the fetus.

[0063] Abnormal skin conductance indicators: When skin conductance is abnormally elevated or the diurnal difference is abnormal, the intervention measures are similar to those for abnormal salivary cortisol, mainly achieved by regulating the pregnant woman's emotions and relieving stress response, including mindfulness meditation, listening to soothing music, and regular work and rest, which can effectively improve the function of the pregnant woman's autonomic nervous system and reduce the risk of premature birth.

[0064] A fetal risk monitoring system based on gestational age physiological characteristics, based on the aforementioned fetal risk monitoring method based on gestational age physiological characteristics, includes a data acquisition module, a data preprocessing module, a risk prediction module, an intervention guidance module, and a closed-loop monitoring module. The functions of each module are as follows:

[0065] Data acquisition module: Used to control various non-invasive home-use data acquisition devices to achieve real-time acquisition, storage and transmission of physiological indicator data. It can interface with saliva collectors, gut microbiota collection boxes, intelligent gait acquisition devices, skin conductance acquisition patches, etc. It can automatically remind pregnant women to collect indicators according to preset gestational week segments to ensure the timeliness and completeness of the collected data. At the same time, it transmits the collected raw data to the data preprocessing module in real time to support subsequent processing.

[0066] The data preprocessing module receives raw physiological indicator data transmitted from the data acquisition module, performs preprocessing steps such as outlier removal, time-series smoothing, data standardization, and gestational age synchronization alignment, eliminates noise interference, standardizes data format, and outputs high-quality time-series datasets for transmission to the AI ​​risk prediction module; at the same time, it stores the preprocessed data in the system database for easy subsequent querying and tracing.

[0067] The AI ​​risk prediction module receives time-series datasets from the data preprocessing module, runs a multimodal AI model (time-series model + correlation model), and fuses the outputs of the two models through an attention mechanism to obtain the final fetal risk score and corresponding risk level, while predicting the expected time of risk occurrence. The risk score, risk level, and expected risk time are then transmitted to the intervention guidance module and mobile terminal APP to provide a basis for the formulation of intervention measures.

[0068] Intervention guidance module: This module receives risk levels and abnormal physiological indicators from the AI ​​risk prediction module. Based on a pre-set intervention strategy library, it pushes targeted non-invasive intervention measures, including intervention methods, operation steps, execution frequency, and precautions. It also tracks the implementation of intervention measures in real time and reminds pregnant women to complete the intervention on time. The intervention strategy library can be dynamically optimized based on updates to clinical data to ensure the effectiveness of the intervention measures.

[0069] Closed-loop monitoring module: It is used to continuously receive post-intervention physiological indicator data transmitted by the data acquisition module, repeat the data preprocessing and risk prediction steps, dynamically update the fetal risk score and risk level, and adjust the intervention measures according to the risk changes. If the risk score drops to a low-risk level, the monitoring frequency can be appropriately reduced. If the risk score rises, the pregnant woman can be promptly reminded to strengthen the intervention or seek medical treatment, forming a closed-loop management of "monitoring-prediction-intervention-re-monitoring".

[0070] Furthermore, it also includes a monitoring equipment kit and a mobile terminal APP. The mobile terminal APP is used to display monitoring data, risk scores and intervention suggestions in real time, and is used to connect with medical diagnosis and treatment systems to achieve data linkage.

[0071] The technical features not described in detail in this solution are based on the conventional operation and general understanding of those skilled in the art and are derived from existing technologies, and will not be elaborated further here.

[0072] The above are merely exemplary embodiments of the present invention and do not limit the patent scope of the present invention. All equivalent transformations made using the contents of the present invention specification and drawings under the technical concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for monitoring fetal risk based on gestational age physiological characteristics, characterized in that, Includes the following steps: Collect physiological data of pregnant women during pregnancy; The physiological index data are preprocessed; A multimodal model is established, and the preprocessed physiological index data is input into the multimodal model. The multimodal model is used to output the abnormality type of physiological index, as well as the risk score and risk level of the fetus. Based on the risk level and the type of abnormality in the physiological indicators, implement targeted non-invasive intervention measures; Continuously monitor physiological indicators after intervention and dynamically update risk assessment results.

2. The fetal risk monitoring method based on gestational age physiological characteristics according to claim 1, characterized in that, The physiological indicators include salivary hormone levels, gut microbiota levels, gait parameters, and skin conductance parameters. These physiological indicators are collected in a targeted manner according to different stages of gestation, which correspond to key stages of fetal development.

3. The fetal risk monitoring method based on gestational age physiological characteristics according to claim 1, characterized in that, The physiological data are collected using a non-invasive home-use collection device, which includes a saliva collector, an intestinal flora collection box, an intelligent gait collection device, and a skin conductance collection patch. The collection device is used to collect data in real time and synchronize the data to a data processing terminal.

4. The fetal risk monitoring method based on gestational age physiological characteristics according to claim 1, characterized in that, The data preprocessing includes removing outliers from physiological indicator data, eliminating short-term data fluctuations through time-series smoothing, normalizing the obtained data, and finally aligning the processed data with gestational age.

5. The fetal risk monitoring method based on gestational age physiological characteristics according to claim 1, characterized in that, The multimodal model includes a temporal model and an association model. The temporal model is used to capture the temporal variation characteristics of physiological indicators and output a temporal risk score. The association model is used to mine the implicit associations between physiological indicators and output an association risk score. The temporal risk score and the association risk score are fused through an attention mechanism to obtain the abnormality type of physiological indicators and the final fetal risk score.

6. The fetal risk monitoring method based on gestational age physiological characteristics according to claim 5, characterized in that, The attention mechanism fusion adopts a preset weight allocation method, which is determined through clinical data calibration, and the sum of the weights of the time-series risk score and the associated risk score is 1.

7. The fetal risk monitoring method based on gestational age physiological characteristics according to claim 6, characterized in that, The fetal risk score is combined with a preset threshold, and the fetal risk score is divided into multiple levels to obtain the fetal risk level. Different fetal risk levels correspond to different monitoring methods and intervention strategies.

8. The fetal risk monitoring method based on gestational age physiological characteristics according to claim 1, characterized in that, The targeted non-invasive intervention measures correspond one-to-one with the types of abnormal physiological indicators.

9. A fetal risk monitoring system based on gestational age physiological characteristics, based on the fetal risk monitoring method based on gestational age physiological characteristics as described in claims 1-8, characterized in that, It includes a data acquisition module, a data preprocessing module, a risk prediction module, an intervention guidance module, and a closed-loop monitoring module.

10. The fetal risk monitoring system based on gestational age physiological characteristics according to claim 9, characterized in that, It also includes a monitoring equipment kit and a mobile terminal APP. The mobile terminal APP is used to display monitoring data, risk scores and intervention suggestions in real time, and to connect with medical diagnosis and treatment systems to achieve data linkage.

Images