Perinatal emotional health monitoring management method and system based on multi-dimensional data association

Abstract

The application discloses a perinatal emotional health monitoring management method and system based on multi-dimensional data association, and relates to the technical field of perinatal female emotional health treatment management. The application realizes the dynamic prediction of emotional risks by collecting pregnant and postpartum data and infant physical sign data, identifying the time lag association and trend association between sleep quality, social support, diet nutrition, exercise state and emotional state through correlation analysis during pregnancy; the infant physical sign data is included in the correlation graph during postpartum, the data correlation graph of the mother and the infant is constructed, the lagging influence of infant physical sign abnormalities on the mother's emotion is identified, and the dynamic assessment of the mother's postpartum emotional risks is realized; and the personalized multi-modal composite intervention scheme is matched and generated from the multi-modal intervention content library based on the correlation analysis result, and the intervention strategy is dynamically optimized. The application realizes the standardized assessment, multi-dimensional data correlation analysis and precise personalized intervention of perinatal female emotional health.

Description

Technical Field

[0001] This invention relates to the field of perinatal women's emotional health treatment and management technology, and in particular to a method and system for perinatal emotional health monitoring and management based on multidimensional data correlation. Background Technology

[0002] Perinatal depression refers to mild to moderate depressive symptoms in women during pregnancy and one year postpartum, including prenatal depression and postpartum depression. It is one of the most common mental disorders during pregnancy and the puerperium. In recent years, the development of mobile health technology has provided new avenues for perinatal mood management. Many software programs now offer features such as mood tracking and health knowledge updates, which have improved the accessibility and convenience of perinatal health management to some extent. However, these existing software programs still have the following shortcomings.

[0003] (1) Disconnect between scale assessment and daily behavior data: The existing perinatal health management system fails to effectively integrate standardized scale assessment with users’ daily recorded dietary, exercise and other behavioral data, resulting in a lack of continuity and comprehensiveness in the assessment results.

[0004] (2) The correlation between the data of each dimension has not been explored: Although it is known that sleep quality, social support, diet and nutrition, exercise status and infant health status are all factors that affect perinatal emotions, the existing technology has failed to conduct correlation analysis on the data of each dimension.

[0005] (3) Postpartum maternal health management and infant health monitoring are separated: In the existing system, maternal health management and infant health monitoring are usually two separate modules, and the infant's physical signs data are not included in the maternal emotional risk assessment system.

[0006] (4) Lack of effective closed-loop feedback mechanism: Most existing systems are one-way information push, which fail to track the user's intervention implementation and effect feedback, and fail to dynamically optimize intervention strategies based on feedback data.

[0007] Therefore, there is an urgent need for a perinatal emotional health monitoring and management method that can deeply integrate standardized scale assessments with daily behavioral data, perform correlation analysis on health data across various dimensions, incorporate postpartum infant physical signs into maternal emotional risk assessment, and possess closed-loop feedback optimization capabilities, so as to achieve dynamic and accurate assessment and personalized intervention of perinatal depression risk. Summary of the Invention

[0008] This invention provides a method and system for monitoring and managing perinatal emotional health based on multidimensional data correlation. By organically integrating maternal pregnancy and postpartum data with infant vital signs data, it achieves continuous and dynamic assessment of perinatal emotions. It performs correlation analysis on multidimensional data such as sleep quality, social support, diet and nutrition, exercise status, and infant vital signs; constructs a "mother-infant" dual-subject data correlation map, thereby incorporating postpartum infant vital signs data into the maternal emotional risk assessment system; and ultimately generates personalized multimodal composite intervention plans, achieving continuous optimization of the intervention plans through closed-loop feedback. This invention is specifically implemented through the following technologies.

[0009] The perinatal emotional health monitoring and management method based on multidimensional data correlation provided by this invention includes the following steps:

[0010] Collect maternal pregnancy and postpartum data and infant vital signs data, perform preprocessing and temporal feature extraction, and construct corresponding temporal feature vectors;

[0011] Align the temporal feature vectors of the pregnancy and postpartum data and the infant vital signs data along the time axis, establish the association edges between the pregnancy and postpartum data and the infant vital signs data, and construct a data association map of mother and infant;

[0012] Based on the data association map, the association between emotional state data in the pregnancy and postpartum data and other pregnancy and postpartum data and infant vital signs data is identified, and a multidimensional association rule set is generated.

[0013] Based on real-time collected pregnancy and postpartum data and infant vital signs data, as well as the multidimensional association rule set, the perinatal impact risk assessment of the mother is carried out, and the emotional risk assessment results of the mother are obtained.

[0014] A multimodal intervention content library is constructed. Based on real-time collected pregnancy and postpartum data, infant vital signs data, and emotional risk assessment results, the current problem dimensions and the hit results in the multidimensional association rule set are identified. Personalized multimodal composite intervention plans are generated by matching the multimodal intervention content library.

[0015] Track the mother's implementation of the multimodal intervention program and her emotional changes to dynamically optimize subsequent intervention strategies.

[0016] Furthermore, the pregnancy and postpartum data include emotional state data, sleep quality data, social support data, dietary and nutritional data, and exercise data during and after pregnancy and postpartum.

[0017] The infant's vital signs data include the infant's heart rate and respiratory rate.

[0018] Furthermore, the emotional state data was collected using a perinatal mood self-assessment scale; the sleep quality data was collected using a sleep quality assessment scale; the social support data was collected using a perinatal support assessment scale; the dietary nutrition data and exercise data were obtained through manual input; and the infant vital signs data were obtained through both manual input and automatic collection.

[0019] Furthermore, the preprocessing method is as follows: imputing missing values, detecting and correcting outliers in the pregnancy and postpartum data and infant vital signs data; and normalizing the pregnancy and postpartum data and infant vital signs data according to their respective dimensions.

[0020] Furthermore, the time-series features extracted from pregnancy and postpartum data and infant vital signs data are as follows:

[0021] (1) EPDS emotional state characteristics: time series of EPDS total score, time series of scores of each item, trend slope of total score (calculated by linear fitting), variance of total score, and extreme frequency of scores of each item.

[0022] (2) PSQI sleep quality characteristics: time series of scores for 7 dimensions and total score, trend slope of each dimension, and variance of each dimension;

[0023] (3) Social support characteristics of MSPSS: time series of scores and total scores in three dimensions, trend slope of each dimension, and variance of each dimension;

[0024] (4) Dietary and nutritional characteristics: time series of daily calorie intake, time series of the proportion of the three major nutrients, trend slope of calorie intake, and dietary regularity index (standard deviation of meal times).

[0025] (5) Exercise characteristics: exercise frequency (number of exercise days per week), average exercise duration, trend slope of exercise duration, and distribution of exercise intensity;

[0026] (6) Infant physical characteristics (postpartum): time series of mean heart rate, heart rate variability, time series of mean respiratory rate, frequency of abnormal physical signs (number of times / day that exceed the normal range).

[0027] Furthermore, the associated edges include time-lag associated edges, time-synchronous associated edges, and trend-related associated edges.

[0028] Furthermore, the time lag correlation edge includes: taking the mother's emotional state data at the current moment as the dependent variable and other pregnancy and postpartum data and infant vital signs data from the past preset time as independent variables, calculating the correlation coefficient between the emotional state data and other pregnancy and postpartum data and infant vital signs data from the past multiple preset time periods (e.g., 1 day ago, 3 days ago, 1 week ago, 2 weeks ago, etc.), and determining the optimal lag time window for the influence of other pregnancy and postpartum data and infant vital signs data on the emotional state data;

[0029] The simultaneous correlation edge includes: within the same time window, calculating the correlation coefficient between the emotional state data and other pregnancy and postpartum data, and infant vital signs data;

[0030] The trend correlation edge includes: calculating the correlation coefficient between the changes in the pregnancy and postpartum data and the infant vital signs data within a preset time window.

[0031] Furthermore, based on the data association graph, the method for identifying the associations between emotional state data in the pregnancy and postpartum data and other pregnancy and postpartum data, as well as infant vital sign data, and generating a multidimensional association rule set is as follows:

[0032] The feature values ​​of pregnancy and postpartum data and infant vital signs data are discretized into different level labels according to preset thresholds to form a level label sequence of each dimension. The FP-Growth algorithm is used to perform frequent pattern mining on the level label sequence to obtain several association rules. The association rules are then filtered and sorted according to support and confidence to generate a multidimensional association rule set.

[0033] Furthermore, based on real-time collected pregnancy and postpartum data and infant vital signs data, as well as the multidimensional association rule set, the method for assessing the perinatal impact risk of mothers and obtaining the emotional risk assessment results of mothers includes: using the current pregnancy and postpartum data and infant vital signs data, as well as the association rule matching features extracted from the multidimensional association rule set as the assessment basis, determining and outputting the current emotional risk assessment results of the user.

[0034] The emotional risk assessment results include the emotional risk level and the emotional risk trend within a preset future time window.

[0035] Furthermore, the multimodal intervention content library includes cognitive behavior guidance content, soothing audiovisual content, lifestyle and support system adjustment suggestions, and maternal-infant interaction guidance.

[0036] Furthermore, the lifestyle and support system adjustment recommendations include sleep hygiene education content, family communication promotion guidelines, social support activation recommendations, personalized nutrition recommendations, and personalized exercise programs.

[0037] Furthermore, the mother-infant interaction guidance includes suggestions for parent-child interaction and guidance on soothing techniques based on infant vital signs data.

[0038] Furthermore, the method for tracking the mother's implementation of the multimodal intervention program and changes in her emotions, and dynamically optimizing subsequent intervention strategies, is as follows:

[0039] Record the mother's viewing, completion, and execution of various intervention components in the multimodal composite intervention plan as process feedback;

[0040] Collect pregnancy and postpartum data and infant vital signs data after the intervention of the multimodal composite intervention program as feedback on the effect;

[0041] Set coefficient thresholds; calculate the actual effectiveness coefficients of various intervention contents in the multimodal composite intervention program based on the process feedback and effect feedback;

[0042] Intervention content with an actual effectiveness coefficient lower than a preset threshold is marked as inefficient content and its matching weight is reduced; intervention content with an actual effectiveness coefficient higher than a preset threshold is marked as efficient content and its matching weight is increased.

[0043] The present invention also provides a perinatal emotional health monitoring and management system based on multidimensional data association, for performing the above-mentioned perinatal emotional health monitoring and management method;

[0044] The perinatal emotional health monitoring and management system includes:

[0045] The multidimensional data acquisition and processing module is used to collect data on the mother's pregnancy and postpartum period.

[0046] The IoT data access module is used to receive infant vital sign data collected by IoT radar sensors and perform empty room filtering and binding verification.

[0047] The temporal feature extraction module is used to preprocess and extract temporal features from the pregnancy and postpartum data and infant vital signs data, and construct the corresponding temporal feature vector.

[0048] The data association graph construction module is used to align the time-series feature vectors of the pregnancy and postpartum data and the infant vital signs data along the time axis, establish the pregnancy and postpartum data and the infant vital signs data, and construct a data association graph between the mother and the infant.

[0049] The association rule mining module is used to identify the associations between emotional state data in the pregnancy and postpartum data and other pregnancy and postpartum data and infant vital signs data based on the data association graph, and generate a multidimensional association rule set.

[0050] The perinatal emotional risk assessment module is used to assess the perinatal impact risk of mothers based on real-time collected pregnancy and postpartum data and infant vital signs data, as well as the multidimensional association rule set, and to obtain the emotional risk assessment results of mothers.

[0051] The personalized intervention module is used to build a multimodal intervention content library. Based on real-time collected pregnancy and postpartum data, infant vital signs data, and emotional risk assessment results, it identifies the current problem dimensions and the hit results in the multidimensional association rule set, and generates personalized multimodal composite intervention plans from the multimodal intervention content library. It tracks the mother's implementation of the multimodal composite intervention plan and emotional changes, and dynamically optimizes subsequent intervention strategies.

[0052] Compared with the prior art, the advantages of the present invention are:

[0053] 1. Deep integration of standardized scales and daily behavior data: This invention integrates three internationally recognized standardized scales, EPDS, PSQI, and MSPSS, with dietary and exercise behavior data manually entered by users every day, realizing a dual-track approach of "standardized assessment + continuous monitoring" for perinatal emotional health, taking into account both the clinical reliability of the assessment and the real-time nature of the monitoring.

[0054] 2. Deep correlation analysis between multi-dimensional data: This invention establishes three types of correlation edges between five dimensions—sleep quality, social support, diet and nutrition, exercise status and infant physical signs—and emotional state: synchronous correlation, time lag correlation, and trend correlation. This can accurately identify the optimal time lag window for each dimension to affect emotions and reveal the multi-factor transmission path of perinatal depression risk.

[0055] 3. Innovative integration of the "mother-infant" dual subject: This invention incorporates infant vital sign data (heart rate, respiration) automatically collected by IoT radar sensors into the postpartum mother's emotional risk assessment system. By constructing a "mother-infant" dual subject data association map, it identifies the delayed impact of abnormal infant vital signs on the mother's emotions, thus upgrading the postpartum depression risk assessment from a "single subject" to a "dual subject".

[0056] 4. Personalized intervention matching based on association rules: Based on the association rule hit rate of data in various dimensions, this invention intelligently matches and combines four intervention modalities to generate personalized intervention plans, realizing the precise strategy of "intervening in whoever affects emotions".

[0057] 5. Closed-loop feedback-driven continuous optimization: This invention establishes a process feedback and effect feedback mechanism based on EPDS retesting and behavioral data changes. By dynamically adjusting the matching weight of intervention content through the effectiveness coefficient, the continuous iterative optimization of the intervention strategy is achieved. Attached Figure Description

[0058] Figure 1 Here is a flowchart of the perinatal emotional health monitoring and management method of the present invention.

[0059] Figure 2 This is a schematic diagram illustrating the method for constructing a data association map between mothers and infants according to the present invention. Detailed Implementation

[0060] The technical solution of the present invention will be clearly and completely described below. 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.

[0061] Example 1

[0062] This invention provides a system for monitoring and managing perinatal emotional health, specifically including the following modules.

[0063] The multidimensional data acquisition and processing module is used to collect data on the mother's pregnancy and postpartum period.

[0064] The IoT data access module is used to receive infant vital sign data collected by IoT radar sensors and perform empty room filtering and binding verification.

[0065] The temporal feature extraction module is used to preprocess and extract temporal features from the pregnancy and postpartum data and infant vital signs data, and construct the corresponding temporal feature vector.

[0066] The data association graph construction module is used to align the temporal feature vectors of the pregnancy and postpartum data and the infant vital signs data along the time axis, establish the association edges between the pregnancy and postpartum data and the infant vital signs data, and construct a data association graph between mother and infant.

[0067] The association rule mining module is used to identify the associations between emotional state data in the pregnancy and postpartum data and other pregnancy and postpartum data and infant vital signs data based on the data association graph, and generate a multidimensional association rule set.

[0068] The perinatal emotional risk assessment module is used to assess the perinatal impact risk of mothers based on real-time collected pregnancy and postpartum data and infant vital signs data, as well as the multidimensional association rule set, and to obtain the emotional risk assessment results of mothers (generally including three levels: high risk, medium risk and low risk).

[0069] The personalized intervention module is used to build a multimodal intervention content library. Based on real-time collected pregnancy and postpartum data, infant vital signs data, and emotional risk assessment results, it identifies the current problem dimensions and the hit results in the multidimensional association rule set, and generates personalized multimodal composite intervention plans from the multimodal intervention content library. It tracks the mother's implementation of the multimodal composite intervention plan and emotional changes, and dynamically optimizes subsequent intervention strategies.

[0070] The system described above employs a three-tier architecture consisting of a server, an administration module, and a mobile client. The server is built using the Spring Boot framework, and its data interaction with the administration module (Vue 2 + Element UI) and the mobile client (uni-app + Vue 3) is achieved through a front-end / back-end separation mechanism.

[0071] The server-side architecture employs a Maven multi-module project structure, divided into an entry module, a general tools module, a security framework module, a business logic module, a scheduled task module, and a code generation module. The server exposes two sets of APIs on the same port: the management API uses JWT token authentication and RBAC access control; the mobile user API uses a custom session token authentication based on WeChat login.

[0072] Example 2: A Method for Monitoring and Managing Perinatal Emotional Health Based on Multidimensional Data Association

[0073] This embodiment utilizes the system described in Embodiment 1 and provides a corresponding perinatal emotional health monitoring and management method based on multidimensional data association, such as... Figure 1 As shown, it can cover the entire perinatal period (i.e., from the beginning of pregnancy to one year postpartum). The specific steps are as follows.

[0074] Step 1: Collect the mother's pregnancy and postpartum data and the baby's vital signs data, perform preprocessing and temporal feature extraction, and construct the corresponding temporal feature vector.

[0075] Depending on the user's stage, pregnancy and postpartum data are divided into individual vital sign data for the mother during pregnancy and postpartum. In this embodiment, the system collects users' health time-series data across various dimensions through a mobile client, and then preprocesses and extracts time-series features from the collected pregnancy and postpartum data and infant vital sign data to construct time-series feature vectors for each dimension. The details are as follows.

[0076] 1. Collection of individual vital signs data during pregnancy (from confirmation of pregnancy to delivery) and postpartum (from delivery to one year postpartum).

[0077] During pregnancy (from confirmation of pregnancy to delivery), the system collects the user's five-dimensional health data in the following manner, and records the time series of scores for each item and the total score:

[0078] (1) Collection of emotional state data

[0079] Data was collected using a perinatal mood self-assessment scale (such as the Edinburgh Postnatal Depression Scale, EPDS). Users completed an EPDS assessment every two weeks.

[0080] The Perinatal Mood Self-Assessment Scale assesses 10 items: mood, pleasure, self-blame, anxiety, fear, coping ability, sleep disturbance, sadness, crying, and self-harming thoughts. Each item is rated on a four-point scale (0-3), with a total score ranging from 0 to 30. Higher scores indicate more severe depressive symptoms. The 10 items are detailed below:

[0081] A. I can see the funny side of things and laugh happily.

[0082] B. I eagerly anticipate everything in the future.

[0083] C. When things go wrong, I unnecessarily blame myself.

[0084] D. I feel anxious and worried for no reason.

[0085] E. I felt fear and panic for no reason.

[0086] F. So many things are coming at me, making it hard for me to breathe.

[0087] G. I am so unhappy that I can't sleep.

[0088] H. I feel sad and heartbroken.

[0089] I'm so unhappy I could cry.

[0090] J. I've thought about hurting myself.

[0091] (2) Sleep quality data collection

[0092] Data is collected using sleep quality assessment scales (such as the Pittsburgh Sleep Quality Index, PSQI). Users complete a PSQI assessment once a month.

[0093] The Sleep Quality Assessment Scale evaluates seven dimensions: subjective sleep quality, sleep latency, sleep duration, sleep efficiency, sleep disorders, use of hypnotics, and daytime dysfunction; the total score ranges from 0 to 21 points. A higher score indicates poorer sleep quality.

[0094] The PSQI scale assesses seven dimensions: subjective sleep quality (e.g., "Overall, how would you rate your sleep quality over the past month?"), sleep latency (e.g., "How long does it usually take you to fall asleep each night over the past month?"), sleep duration (e.g., "How many hours of sleep did you actually get each night over the past month?"), sleep efficiency (the ratio of sleep duration to time spent in bed), sleep disturbances (e.g., "Have you experienced frequent or early awakenings during the past month?"), use of hypnotics (e.g., "Have you used any medication to help you sleep over the past month?"), and daytime dysfunction (e.g., "Have you felt sleepy or lacking energy during the day due to insufficient sleep over the past month?"). Each dimension is scored from 0 to 3.

[0095] (3) Social support data collection

[0096] The perinatal support assessment scale (such as the Multidimensional Perceived Social Support Scale, MSPSS) is used. Users complete the MSPSS assessment once a month.

[0097] The Perinatal Support Assessment Scale evaluates three dimensions: family support, friend support, and other support. It contains 12 items and uses a 7-point rating scale; the total score ranges from 12 to 84 points. A higher score indicates a higher perceived level of social support.

[0098] The 12 items are as follows:

[0099] A. When I encounter problems, some people (leaders, relatives, colleagues) will appear by my side.

[0100] B. I am able to share joy and sorrow with some people (leaders, relatives, colleagues).

[0101] C. My family can provide me with concrete and practical help.

[0102] D. I can get emotional help and support from my family when needed.

[0103] E. When I am in trouble, some people (leaders, relatives, colleagues) are my true source of comfort.

[0104] F. My friends can really help me.

[0105] G. I can rely on my friends when difficulties arise.

[0106] H. I am able to talk about my problems with my family.

[0107] I. My friends can share my joy and sorrow.

[0108] J. There are certain people in my life (leaders, relatives, colleagues) who care about my feelings.

[0109] K. My family is willing to help me make various decisions.

[0110] L. is able to discuss her problems with her friends.

[0111] (4) Collection of dietary nutrition data

[0112] Users manually enter their daily meal information (including food types and portions, in grams). The system automatically calculates daily calorie intake (kcal / day) and the distribution of the three macronutrients (carbohydrates, protein, and fat) based on a preset food database, forming a dietary nutrition time sequence.

[0113] (5) Motion data acquisition

[0114] Users manually input the type of exercise (e.g., can select from a drop-down list: walking / yoga / swimming / aerobics / other), exercise duration (minutes), and subjective exercise intensity (0-10 level slider) to form an exercise time sequence each day.

[0115] The collection of the above five dimensions of data will continue postpartum (from delivery to one year postpartum).

[0116] 2. Collection of infant vital signs data.

[0117] Infant vital signs data include two sources: (1) Manual input by the user: The user manually inputs the infant's heart rate (beats / min) and respiratory rate (breaths / min); (2) Automatic acquisition by oT radar sensor: Non-contact continuous monitoring of infant's heart rate and respiration is achieved through millimeter-wave radar technology.

[0118] The radar sensor reports vital signs data to the server in real time via the MQTT protocol. The server connects to the MQTT Broker via bidirectional TLS (mTLS) authentication, performs empty room filtering (discarding data when both heart rate and respiration are 0) and binding verification (querying the device-user binding relationship in Redis based on the device ID, discarding data if not bound), and writes the verified data to the Redis cache.

[0119] The criteria for judging abnormal signs in infants are as follows: (1) The normal range of heart rate for newborns aged 0-28 days is 100-160 beats / min. If it exceeds this range, it is considered an abnormality. (2) The normal range of respiratory rate for newborns aged 0-28 days is 40-60 breaths / min. If it exceeds this range, it is considered an abnormality. (3) The normal range of heart rate for infants aged 1-12 months is 90-140 beats / min. If it exceeds this range, it is considered an abnormality. (4) The normal range of respiratory rate for infants aged 1-12 months is 30-50 breaths / min. If it exceeds this range, it is considered an abnormality.

[0120] The system counts the total number of abnormal infant vital signs on a daily basis. When the total number of abnormalities on a day is ≥4, it is marked as "significant abnormality in infant vital signs".

[0121] The working principle of the above technical solution is as follows: It achieves systematic collection of multidimensional perinatal health data through mobile user terminals. Data for each dimension has different collection frequencies (EPDS every 2 weeks, PSQI monthly, MSPSS monthly, diet and exercise daily, and infant vital signs in real time), forming a multi-timescale time-series dataset. The postpartum model adds IoT radar sensor data to the five dimensions of pregnancy data, achieving unified collection and timeline alignment of maternal and infant data.

[0122] The effects of the above technical solution are as follows: A phased, multi-dimensional data collection approach provides a complete data foundation for subsequent temporal feature extraction and correlation analysis. Periodic retesting of EPDS can capture the time-varying trajectory of emotions; daily diet and exercise records compensate for the limitations of scale assessments in terms of temporal resolution; IoT radar sensors enable non-contact continuous monitoring of infant vital signs; and unified collection of maternal and infant data provides the data prerequisite for constructing a data correlation map between mothers and infants.

[0123] 3. Preprocessing of pregnancy and postpartum data and infant vital signs data.

[0124] (1) Imputation of missing values

[0125] For occasional missing items in assessments of scales such as EPDS, PSQI, and MSPSS (when the number of missing items on a single scale does not exceed 20% of the total number of items on that scale), the mean of the two assessments is used for imputation.

[0126] For occasional omissions in daily records such as diet and exercise (when there is a missing record on a single day but there are records within 3 days before and after), the forward imputation method is used for imputation.

[0127] Data segments that are missing for more than 7 consecutive days are marked as invalid data segments and are not included in time series feature extraction.

[0128] (2) Outlier detection and correction

[0129] The normal range for heart rate data of infants aged 1-12 months is 90-140 beats / min. Values ​​outside this range are marked as abnormal and corrected by averaging the adjacent valid values.

[0130] The normal range for respiratory rate data is 30-50 breaths / min. Values ​​exceeding this range are marked as abnormal and corrected.

[0131] The total score for EPDS ranges from 0 to 30 points. Data outside this range is marked as invalid and discarded.

[0132] The PSQI total score ranges from 0 to 21 points. Data outside this range is marked as invalid and discarded.

[0133] (3) Data normalization

[0134] The pregnancy and postpartum data and infant vital signs data were standardized using Z-score according to their respective dimensions, so that the mean of each dimension of the data was 0 and the standard deviation was 1.

[0135] 4. Temporal feature extraction.

[0136] For the preprocessed pregnancy and postpartum data and infant vital signs data, the following time-series features were extracted using a 30-day time window and a 7-day sliding step:

[0137] (1) EPDS sentiment characteristics: EPDS total score mean, score mean of each item, trend slope of total score (calculated by least squares linear fitting), variance of total score fluctuation, and percentage of each item with a score ≥2.

[0138] (2) PSQI sleep characteristics: mean scores of 7 dimensions and mean total score, trend slope of each dimension, and variance of each dimension.

[0139] (3) Social support characteristics of MSPSS: mean scores of three dimensions and mean total score, trend slope of each dimension, and variance of each dimension.

[0140] (4) Dietary and nutritional characteristics: average daily calorie intake (kcal / day), average proportion of the three major nutrients, trend slope of calorie intake, and standard deviation of meal times (dietary regularity index, unit: minutes).

[0141] (5) Exercise characteristics: average number of exercise days per week, average duration of a single exercise session (minutes), trend slope of exercise duration, and average exercise intensity (0-10 levels).

[0142] (6) Infant physical characteristics (postpartum): mean heart rate (beats / min), heart rate variability (standard deviation, unit: beats / min), mean respiratory rate (beats / min), frequency of abnormal physical signs (number of times / day that exceed the normal range).

[0143] 5. Construction of temporal feature vectors.

[0144] The time-series feature values ​​extracted from the above dimensions are arranged in order according to the predefined feature index to form a fixed-length time-series feature vector, which serves as the data input for subsequent association map construction and perinatal emotional risk assessment.

[0145] The working principle of the above technical solution is as follows: This embodiment solves the problem of incompleteness in real data collection through preprocessing steps; it transforms the original pregnancy and postpartum data and infant vital signs data into calculable numerical features through statistical measures such as trend slope and fluctuation variance; and it unifies multi-source heterogeneous data into a standardized format through time-series feature vectors.

[0146] The effects of the above technical solutions are as follows: missing value imputation and outlier correction improve data quality; features such as trend slope and variance of fluctuation transform discrete observations into dynamic indicators that reflect changing trends; and the construction of time series feature vectors provides standardized input for the FP-Growth algorithm and risk assessment.

[0147] Step 2: Align the time-series feature vectors of the pregnancy and postpartum data and the infant vital signs data along the time axis, establish the association edges between the pregnancy and postpartum data and the infant vital signs data, and construct a data association graph between the mother and the infant.

[0148] In this embodiment, the construction of the data association map is divided into two stages: pregnancy and postpartum.

[0149] 1. Construction of a pregnancy association map

[0150] (1) Node definition

[0151] The nodes in the data association graph include five categories: "Mother's Emotional State Node" (EPDS total score and 10 item dimensions), "Mother's Sleep Quality Node" (PSQI 7 dimensions), "Mother's Social Support Node" (MSPSS 3 dimensions and 12 items), "Mother's Diet and Nutrition Node" (calorie intake, percentage of the three macronutrients), and "Mother's Exercise Node" (exercise frequency, duration, and intensity). Each node carries a corresponding temporal feature vector.

[0152] (2) Time axis alignment

[0153] The mother's EPDS, PSQI, and MSPSS assessment data were aligned with her daily diet and exercise records along a timeline to form a multidimensional time-series dataset for the entire pregnancy, organized by week.

[0154] In cases where the assessment date differs from the daily record date, the average of the data for each dimension within three days before and after the assessment date is taken as the alignment value for that assessment time point.

[0155] (3) Construction of edges associated with the same time

[0156] Within the same time window, the correlation coefficients between the EPDS score and other pregnancy and postpartum data and infant vital signs data were calculated, i.e., the "emotional state-sleep quality contemporaneous association", "emotional state-social support contemporaneous association", "emotional state-dietary contemporaneous association", and "emotional state-exercise contemporaneous association" were constructed.

[0157] Within the same time window, calculate the Pearson correlation coefficient between the EPDS total score and the aforementioned pregnancy and postpartum data for each other dimension. The calculation formula is as follows:

[0158] ;

[0159] Where x is the EPDS total score, and y is the other dimensions of pregnancy and postpartum data (PSQI total score, MSPSS total score, average daily calorie intake, and number of exercise days per week). The correlation coefficients r1 (emotional state - sleep quality), r2 (emotional state - social support), r3 (emotional state - diet), and r4 (emotional state - exercise) were calculated.

[0160] (4) Construction of time-lag related edges

[0161] Using the current EPDS score (mother's emotional state data) as the dependent variable and other pregnancy and postpartum data and infant vital signs data from preset past times (such as 1 day ago, 3 days ago, 1 week ago, 2 weeks ago, etc.) as independent variables, the correlation coefficients between the other pregnancy and postpartum data and infant vital signs data and the emotional state data at each time point are calculated to determine the optimal lag window for the influence of the other pregnancy and postpartum data and infant vital signs data on emotions.

[0162] ① Sleep lag association: Calculate the correlation coefficients between the PSQI score and the current EPDS score at 1, 2, 3, and 4 weeks lag, and take the lag number T corresponding to the maximum value. sleep Sleep serves as the best delayed window for influencing mood.

[0163] ②Lagged Association of Social Support: Calculate the correlation coefficients between the MSPSS score and the current EPDS score at lags of 1 week, 2 weeks, 3 weeks, and 4 weeks, and take the lag number T corresponding to the maximum value. support As the best lagging window for the influence of social support on emotions.

[0164] ③Lapse correlation of diet: Calculate the correlation coefficients between the mean calorie intake lagged by 3 days, 7 days, and 14 days and the current EPDS score, and take the lag number T corresponding to the maximum value. diet As the best lagging window for the influence of diet on mood.

[0165] ④ Movement lag correlation: Calculate the correlation coefficients between movement frequency and the current EPDS score at lags of 3 days, 7 days, and 14 days, and take the lag number T corresponding to the maximum value. exercise As the best lagging window for the impact of exercise on emotions.

[0166] For example, by comparing the correlation coefficients between the PSQI score at a 1-week lag and the current EPDS score, which are higher than those of the same period, it is found that the effect of sleep on mood has a lag effect of about 1 week. Similarly, we can identify that "changes in social support affect mood scores at a 2-week lag" and "changes in dietary intake affect mood scores at a 3-day lag".

[0167] (5) Construction of trend-related edges

[0168] Calculate the correlation coefficient between the slope of the trend of each of the pregnancy and postpartum data and infant vital signs data within a preset time window (e.g., 90 days), and identify trend association patterns such as "continuous sleep deterioration accompanied by synchronous mood deterioration", "declining social support accompanied by mood deterioration", and "declining eating regularity accompanied by increased mood fluctuations".

[0169] ① Perform linear fitting on the EPDS total score, PSQI total score, MSPSS total score, daily calorie intake, and number of exercise days per week, and calculate the slope k of their respective trends. epds k psqi k mspss k diet k exercise ;

[0170] ② Calculate the correlation coefficient between trend slopes to identify trend association patterns such as "continuous sleep deterioration accompanied by synchronous mood deterioration" and "declining social support accompanied by mood deterioration".

[0171] 2. Construction of postpartum association map

[0172] (1) Add a new node

[0173] In the postpartum stage, based on the aforementioned five-dimensional correlation map, "infant vital signs nodes" (mean heart rate, mean respiratory rate, frequency of abnormal vital signs) are added, carrying their temporal feature vectors, to construct a data correlation map between mother and infant, such as... Figure 2 As shown.

[0174] (2) Add related edges

[0175] Establish three types of association edges between the "infant vital signs node" and each mother node:

[0176] ①Simultaneous correlation: Calculate the correlation coefficient r5 (emotional state - infant physical signs) between the frequency of abnormal infant physical signs on the same day and the mother's EPDS score on the same day.

[0177] ②Lapse correlation: Calculate the correlation coefficients between the frequency of abnormal infant vital signs and the mother's EPDS score on the day of birth for the previous 1, 2, 3, 4, 5, 6, and 7 days, respectively, and take the lag period T corresponding to the maximum value. baby This serves as the optimal lag window for how an infant's physical signs influence their emotions.

[0178] For example, if the correlation coefficient between "abnormal infant physical signs delayed by 2 days" and "mother's mood on the day" is found to be the highest, it indicates that the impact of abnormal infant physical signs on mother's mood has a lag effect of about 2 days.

[0179] ③ Trend association: Calculate the correlation coefficient between the 7-day trend slope of the frequency of abnormal infant vital signs and the 7-day trend slope of the mother's EPDS score to identify the association pattern of "the upward trend of the frequency of abnormal infant vital signs accompanied by the trend of the mother's mood deterioration".

[0180] 3. Storage and updating of association maps

[0181] The association graph is stored in the database in a graph structure, and the weight of each association edge is represented by the correlation coefficient, with a value range of [-1, 1].

[0182] As user data continues to accumulate, the association graph is updated every two weeks of new valid data, and the weight values ​​of each association edge are recalculated.

[0183] The working principle of the above technical solution is as follows: by constructing a correlation graph of six types of nodes, the correlation between data in various dimensions is quantified into a computable graph structure. The three types of correlation edges respectively characterize the correlation at different time dimensions—synchronous correlation reflects coordinated changes; time-lag correlation reveals the best lag window affecting emotions; and trend correlation reflects the synchronicity of long-term change trends.

[0184] The effects of the above technical solution are as follows: the time lag correlation edge can accurately identify the best lag window for each dimension to affect emotions, providing a time anchor for early warning; the correlation edge between the newly added mothers (pregnancy and postpartum data) and infants (infant vital signs data) after childbirth realizes the leap from "single subject" to "dual subject"; the dynamic update of the correlation graph makes the analysis results increasingly consistent with the user's actual situation.

[0185] Step 3: Based on the data association graph, identify the associations between the emotional state data in the pregnancy and postpartum data and other pregnancy and postpartum data and infant vital signs data, and generate a multidimensional association rule set.

[0186] In this embodiment, the system uses an association rule mining algorithm to identify implicit associations between emotional state data in the pregnancy and postpartum data and other pregnancy and postpartum data, as well as infant vital signs data. Specifically, the system mines multidimensional association rules and constructs a multidimensional association rule set through the following steps.

[0187] 1. Data discretization

[0188] The continuous feature values ​​of each dimension are discretized into level labels according to preset clinical thresholds. The specific thresholds are as follows:

[0189] ①EPDS total score: ≥13 points is labeled as “high risk”, 10-12 points is labeled as “medium risk”, and ≤9 points is labeled as “low risk”.

[0190] ②PSQI total score: ≥8 points is the “sleep disorder” label, 5-7 points is the “normal sleep” label, and ≤4 points is the “good sleep” label.

[0191] ③MSPSS total score: ≥60 points is the “sufficient support” label, 37-59 points is the “moderate support” label, and ≤36 points is the “lack of support” label.

[0192] ④ Dietary calorie intake: Below 80% of the recommended intake is labeled "insufficient intake", 80%-120% of the recommended intake is labeled "normal intake", and above 120% is labeled "excessive intake".

[0193] ⑤ Exercise frequency: <3 days / week is labeled "lack of exercise", ≥3 days / week is labeled "sufficient exercise".

[0194] ⑥ Frequency of abnormal infant vital signs (postpartum): 0 times / day is labeled "no abnormality", 1-3 times / day is labeled "mild abnormality", and ≥4 times / day is labeled "significant abnormality".

[0195] As shown in Table 1 below.

[0196] Table 1

[0197]

[0198] 2. Construction of transaction database

[0199] Each user's evaluation scores for each dimension are combined into a single transaction record, formatted as: {user_id, evaluation_date, EPDS level, PSQI level, MSPSS level, diet level, exercise level, infant vital signs level (postpartum)}. All users' transaction data collectively constitute the transaction database.

[0200] 3. Frequent Pattern Mining Using FP-Growth Algorithm

[0201] FP-Growth (Frequent Pattern Growth) is a classic association rule mining algorithm. Its core idea is to compress transaction data by constructing a compact prefix tree structure called "Frequent Pattern Tree (FP-Tree)". It can efficiently find all frequent itemsets by scanning all data only twice, without generating a large number of candidate sets like the Apriori algorithm.

[0202] Specifically, health data features across various dimensions (such as "sleep disorders," "lack of social support," and "high emotional risk") are treated as data items. Each user's assessment results are combined into transaction data, and the FP-Growth algorithm is used to mine all frequently occurring feature combinations. Subsequently, the system calculates the support (i.e., the frequency of the combination in all transactions) and confidence (i.e., the conditional probability that the consequent is also included in all transactions containing the antecedent), and filters them according to preset minimum support and minimum confidence thresholds.

[0203] This embodiment uses the FP-Growth algorithm to mine frequent patterns in the grade label sequence. The algorithm parameters are set as follows: minimum support threshold = 10%, minimum confidence threshold = 70%. The specific mining steps are as follows.

[0204] (1) Scan the transaction database for the first time and count the 1-itemset support of each level label (i.e., the frequency of a single level label in all transactions); filter out data items with support below 10% (e.g., the "normal diet intake" label appears less than 10% of the time and is not included in subsequent mining); sort the remaining data items in descending order of support and construct a frequent item header table.

[0205] (2) Scan the transaction database a second time, arrange the data items in each transaction in the order of the frequent item header table, and insert them into the FP-Tree.

[0206] The insertion rule is as follows: Starting from the root node, if the current node has a child node with the same name as the current level label, then the child node's count is incremented by 1; otherwise, a new child node is created, and its count is 1. Branches on the transaction-shared tree with the same level label combination.

[0207] (3) Recursively mine frequent itemsets from FP-Tree: Starting from the bottom of the frequent itemset header table, construct its conditional pattern base (i.e., the set of all prefix paths containing the level label) for each level label; then recursively construct the conditional FP-Tree on the conditional pattern base, repeat the above process until the conditional FP-Tree is empty, and output all frequent itemsets that meet the minimum support requirement (i.e., frequent combinations of multiple dimension level labels).

[0208] 4. Association rule generation and filtering

[0209] Association rules are generated from the frequently mined item set to form a multidimensional association rule set.

[0210] For each frequent itemset, divide it into antecedent A and consequent B (B being a single item), and calculate the confidence score Confidence(A→B) = Support(A∪B) / Support(A). Retain rules with a confidence score ≥70%.

[0211] The following are examples of several output association rules.

[0212] Rule 1: Sleep disorders + lack of social support → high risk of emotional distress (support level 15%, confidence level 85%).

[0213] Rule 2: Irregular eating habits + lack of exercise → medium risk in mood (support level 12%, confidence level 72%).

[0214] Rule 3: Sleep disturbance (1 week lag) → High risk of emotional distress (18% support, 78% confidence).

[0215] Rule 4: Lack of social support (2-week lag) → High risk of emotional distress (14% support, 80% confidence).

[0216] Rule 5 (postpartum): Significantly frequent abnormal infant signs + insufficient sleep → high risk of maternal emotional distress (support 10%, confidence 88%).

[0217] It should be noted that in practical applications, the association rules are not limited to the above five types and need to be determined based on the actual situation.

[0218] The working principle of the above technical solution is as follows: The FP-Growth algorithm compresses transaction data using an FP-Tree, efficiently identifying all frequent itemsets with only two scans. Data discretization transforms continuous numerical values ​​into rank labels, preserving clinical grading significance. Confidence screening ensures the rules have predictive value.

[0219] The effects of the above technical solutions are as follows: the FP-Growth algorithm improves computational efficiency by tens of times compared with the traditional Apriori algorithm; the dual threshold screening (support ≥10%, confidence ≥70%) ensures the statistical reliability and predictive value of the mining results.

[0220] Step 4: Based on real-time collected pregnancy and postpartum data and infant vital signs data, as well as the multidimensional association rule set, conduct a perinatal risk assessment for the mother and obtain the mother's emotional risk assessment results.

[0221] In this embodiment, the system employs the following method to achieve dynamic assessment of perinatal emotional risks.

[0222] 1. Input Construction for Perinatal Emotional Risk Assessment

[0223] The evaluation inputs are the user's current feature values ​​(EPDS, PSQI, MSPSS, diet, exercise, infant vital signs) and the association rule matching features extracted from the multidimensional association rule set.

[0224] Specifically, the system compares the current level labels of each dimension with each rule in the multidimensional association rule set. If all the conditions in the antecedent (i.e., the level labels of all dimensions to the left of the arrow in step 4 of step 3) are met, the rule is marked as "hit", and the hit rule ID and corresponding confidence value are recorded.

[0225] 2. Current risk level assessment

[0226] The user's current emotional risk level (low risk / medium risk / high risk) is determined based on the following rules:

[0227] (1) If the total EPDS score is ≥13 points, it is directly judged as a high-risk level;

[0228] (2) If the total score of EPDS is 10-12 points and at least one association rule with a confidence level of ≥80% is hit, it is judged as a high-risk level;

[0229] (3) If the total score of EPDS is 10-12 points and no association rule with a confidence level of ≥80% is hit, it is judged as medium risk level;

[0230] (4) If the total EPDS score is ≤9, but 3 or more association rules with a confidence level ≥70% are hit, it is judged as medium risk level;

[0231] (5) The remaining situations are classified as low-risk.

[0232] 3. Future Risk Trend Forecast

[0233] Based on the time lag correlation parameters in the user's personalized data correlation graph, the predicted value of the direction of change of emotional risk within the future preset time window is output.

[0234] The "prediction of emotional risk trends within a future preset time window" refers to the system outputting a predicted value for the direction of change in emotional scores within a fixed future time frame, based on the user's personalized correlation graph, especially the optimal lag time for the impact of data on emotional state in each dimension (such as a 1-week lag for sleep and a 2-day lag for abnormal infant vital signs), combined with the real-time status of current multidimensional data. This predicted value includes three types: "upward trend (deterioration)," "downward trend (improvement)," and "stable trend."

[0235] The specific steps are as follows.

[0236] (1) Extract the optimal lag time parameters for each dimension of the user from the association graph: sleep lag window T sleep (Week), Social Support Lagging Window T support (Week), Dietary Delay Window T diet (day), motion lag window T exercise (days), Postpartum Infant Physical Sign Delay Window (T) baby (sky).

[0237] (2) Obtain the current time data of each antecedent dimension relative to T. sleep / T diet / T baby The direction of change before the time.

[0238] For example, the PSQI score at the current moment compared to T sleep The change in PSQI score from the previous week, ΔPSQI, is marked as "deterioration direction" if ΔPSQI ≥ 2 points (an increase in PSQI indicates worsening sleep); "improvement direction" if ΔPSQI ≤ -2 points; and "stable direction" if -2 < ΔPSQI < 2.

[0239] (3) Calculate the trend prediction value TrendScore by combining the change direction and correlation weight of each antecedent dimension:

[0240] TrendScore=Σ(W i × Dir i );

[0241] Among them, W i Dir is the confidence value (a decimal between 0 and 1) of the i-th association rule. i The trend is the direction of change in the antecedent dimension of the i-th rule (deterioration = +1, improvement = -1, stability = 0). A TrendScore > 0.3 is considered an "upward trend" (increased risk), a TrendScore < -0.3 is considered a "downward trend" (decreased risk), and a TrendScore ≤ -0.3 is considered a "stable trend".

[0242] 4. Regular dynamic assessment and early warning triggering

[0243] (1) The emotional risk assessment process runs automatically each time a user completes an EPDS assessment, outputting the current risk level and future trend; at the same time, a scheduled task is executed at 2:00 a.m. every day to run an emotional risk assessment once for all active users (users who have reported data in the last 7 days).

[0244] (2) Warning Trigger

[0245] When the risk level output by the emotional risk assessment result is "high risk", or the predicted future trend is "upward trend" and the increase exceeds the preset threshold (e.g., TrendScore > 0.5), an early warning will be automatically triggered, and warning information will be pushed to the user and management terminal.

[0246] The working principle of the above technical solution is as follows: risk level is determined by combining EPDS scores and correlation rule hits through a rule-driven method, future sentiment trends are predicted through personalized lagged correlation parameters, and the timeliness of risk assessment is ensured through a regular evaluation mechanism.

[0247] The effects of the above technical solutions are as follows: the risk level assessment combines EPDS scores and correlation rule hits, improving the sensitivity of early warning; the future trend prediction achieves early warning based on personalized lagging correlation parameters; the threshold setting of TrendScore (>0.3 indicates an upward trend, >0.5 triggers an early warning) ensures the accuracy of early warning; and the daily early morning scheduled operation ensures that signals of deteriorating emotions can be captured at the earliest possible time.

[0248] Step 5: Construct a multimodal intervention content library. Based on real-time collected pregnancy and postpartum data, infant vital signs data, and emotional risk assessment results, identify the current problem dimensions and the hit results in the multidimensional association rule set, and generate personalized multimodal composite intervention plans from the multimodal intervention content library.

[0249] In this embodiment, the system generates a personalized multimodal composite intervention plan through the following steps.

[0250] 1. Construct a multimodal intervention content library

[0251] The multimodal intervention content library includes four parts: cognitive behavior guidance content, soothing audiovisual content, lifestyle and support system adjustment suggestions, and maternal-infant interaction guidance (postpartum).

[0252] (1) Cognitive behavioral guidance content: self-guided texts and audio based on the principles of cognitive behavioral therapy, including cognitive restructuring exercises, behavioral activation suggestions, etc.

[0253] (2) Soothing audiovisual content: emotionally soothing animations, guided meditation videos, and relaxing music.

[0254] (3) Lifestyle and support system adjustment recommendations: sleep hygiene education content based on PSQI score, family communication promotion guidelines and social support activation recommendations based on MSPSS score, personalized nutrition recommendations based on diet records, and personalized exercise programs based on exercise records.

[0255] (4) Guidance on mother-infant interaction (postpartum): suggestions for parent-child interaction and guidance on soothing techniques based on infant vital signs data.

[0256] 2. Problem Dimension Identification

[0257] Based on the user's risk level output from the perinatal emotional risk assessment results, as well as the status of pregnancy and postpartum data and infant vital signs data, the specific dimensions of the user's current problems are identified. The judgment criteria are as follows:

[0258] (1) Emotional state dimension: EPDS total score ≥10 points, or EPDS item 10 (self-harming thoughts) score ≥1 point, marked as "emotional problem".

[0259] (2) Sleep quality dimension: PSQI total score ≥8 points, or PSQI sleep latency dimension ≥2 points, or PSQI daytime dysfunction dimension ≥2 points, are marked as "sleep problems".

[0260] (3) Social support dimension: If the MSPSS family support dimension score is ≤12 (out of 28), or the total MSPSS score is ≤36, it is marked as "social support problem".

[0261] (4) Dietary nutrition dimension: If the daily calorie intake is lower than 80% of the recommended amount for more than 3 consecutive days, or if the daily calorie intake is higher than 120% of the recommended amount for more than 3 consecutive days, it is marked as "dietary problem".

[0262] (5) Exercise dimension: If the number of exercise days per week is less than 3 days for more than 2 consecutive weeks, it is marked as "exercise problem".

[0263] (6) Infant physical signs dimension (postpartum): If the frequency of abnormal infant physical signs is ≥4 times / day for more than 2 consecutive days, it is marked as “abnormal infant physical signs”.

[0264] 3. Association rule hit

[0265] Find the association rules that match the current status, sort them from high to low confidence level, and determine the intervention priority for each pregnancy, postpartum data, and infant vital signs data.

[0266] The antecedent dimension corresponding to the rule with the highest confidence is marked as "first-level intervention target", and the antecedent dimension of the other hit rules is marked as "second-level intervention target".

[0267] 4. Multimodal content matching

[0268] From the aforementioned pre-defined multimodal intervention content library, based on the user's risk level output from the perinatal emotional risk assessment results, the status of pregnancy and postpartum data, infant vital signs data, and the results of association rule matching, the optimal content is selected from each modality for combination.

[0269] (1) Content of cognitive behavior guidance

[0270] For users with an EPDS score of ≥10, cognitive behavioral therapy self-guided text and audio will be matched, including cognitive restructuring exercises, behavioral activation task lists, and mindfulness guided audio, with a push frequency of 3 times per week.

[0271] (2) Soothing audiovisual content

[0272] All users will be matched with mood-soothing animations (3-5 minutes long), guided meditation videos (5-10 minutes long), and relaxing music (15-30 minutes long), with a push frequency of once a day.

[0273] (3) Recommendations for lifestyle and support system adjustments

[0274] For users with a PSQI score of 8 or higher, sleep hygiene education content (text + audio, 5-8 minutes reading time) will be provided. For users with an MSPSS score of 36 or lower, guidance on promoting family communication and suggestions for activating social support will be provided (text + interactive questionnaire). For users with dietary or exercise issues, personalized nutrition advice and exercise plans (text + charts) will be provided, with push notifications sent twice a week.

[0275] (4) Guidance on mother-infant interaction (postpartum)

[0276] For users whose infants exhibit abnormal vital signs ≥ 4 times / day, we will provide parent-child interaction suggestions and soothing techniques guidance (video + text, 3-8 minutes in length); including kangaroo care operation instructions, soothing postures and patting techniques demonstrations, with a push frequency of once a day until the abnormal frequency drops to less than once / day.

[0277] Specifically, intervention is prioritized for the dimension corresponding to the association rule with the highest confidence level. For example, if the current state matches "sleep disorder + lack of social support → high risk of emotional distress" (confidence level 85%), then sleep hygiene education content and social support activation suggestions are matched from modality (3) first. If the postpartum rule "significant frequency of abnormal infant physical signs + insufficient sleep → high risk of maternal emotional distress" (support level 10%, confidence level 88%) is matched at the same time, then maternal-infant interaction guidance content is matched from modality (4) in addition.

[0278] 5. Generation and delivery of multimodal composite intervention programs

[0279] Based on the intervention priority and the generated multimodal composite intervention plan, generate a multimodal composite intervention plan containing 2-3 modalities.

[0280] Multimodal combined intervention programs are arranged in order of priority: modal content corresponding to primary intervention targets is arranged first, and modal content corresponding to secondary intervention targets is arranged last.

[0281] The multimodal intervention program is delivered to the user's device via a message interface of a specific software program in a mixed format of text, audio, and video. The push time is set according to user preferences (users can set their preferred receiving time period in the software client, with the default being 20:00-21:00 daily), with no more than two pushes per day, and each push content not exceeding 30 minutes in total duration.

[0282] The working principle of the above technical solution is as follows: accurately locate the problem dimension by determining thresholds of various dimensions, determine the intervention priority by ranking by rule confidence, achieve precise intervention by matching multimodal content library, and generate multimodal composite intervention schemes to meet the diverse needs of users.

[0283] The above technical solutions achieve the following effects: problem dimension identification is based on specific dimension scores of the scale rather than relying solely on the total score, accurately locating the source of the problem; association rule matching ensures priority intervention for factors with the greatest impact on emotions through confidence ranking; multimodal content matching enables the precision of the intervention plan; and fixed parameters for push frequency and duration guarantee the controllability and reproducibility of the system.

[0284] Step Six: Track the mother's implementation of the multimodal intervention program and her emotional changes, and dynamically optimize subsequent intervention strategies.

[0285] In this embodiment, the system achieves closed-loop feedback and continuous optimization of the intervention strategy through the following steps. The closed-loop feedback mechanism includes process feedback collection, effect feedback collection, effectiveness evaluation, and dynamic strategy optimization. By tracking the user's execution of the multimodal composite intervention plan and changes in data across various dimensions through the closed-loop feedback mechanism, subsequent intervention strategies are dynamically optimized.

[0286] 1. Process feedback collection

[0287] The system records user interactions with various interventions within the multimodal composite intervention plan, including viewing, completion, and execution status.

[0288] (1) Open status: Whether the user clicks to open the multimodal composite intervention scheme. Calculate the open rate = number of times opened / number of times pushed.

[0289] (2) Completion status: The completion status of each modality of content. Calculate the completion rate = number of completions / number of opens.

[0290] Specifically, audio content is considered complete when the playback progress is ≥90%, video content is considered complete when the playback progress is ≥95%, and text and image content is considered complete when scrolling to the bottom.

[0291] (3) Implementation status: The implementation status of the lifestyle recommendations is indirectly determined by changes in user behavior data over the next 7 days.

[0292] If, after an exercise recommendation is sent, the user's exercise frequency increases by ≥1 day / week in the following 7 days compared to the 7 days prior to the recommendation, it is marked as "Executed". If, after a sleep hygiene education message is sent, the user's PSQI sleep latency score decreases by ≥1 point in the following 7 days, it is marked as "Executed". If, after an infant care plan is sent, the frequency of ≥4 abnormal vital signs per day decreases by ≥2 days in the following 7 days compared to the 7 days prior to the recommendation, it is marked as "Care Executed" (applicable postpartum).

[0293] 2. Collection of Results Feedback

[0294] The system automatically and continuously collects the following data on the 14th day (±2 days) after the multimodal composite intervention plan is pushed out.

[0295] (1) EPDS retest data: Record the total EPDS score and change before and after intervention ΔEPDS = EPDS after - EPDS before (negative value indicates improvement).

[0296] (2) PSQI retest data: Record the total PSQI score and change ΔPSQI before and after intervention (negative values ​​indicate improvement).

[0297] (3) MSPSS retest data: Record the total MSPSS score and change ΔMSPSS before and after the intervention (positive value indicates improvement).

[0298] (4) Changes in dietary behavior: Record the change in the rate of calorie intake compliance (the percentage of days falling within 80%-120% of the recommended intake) ΔDiet before and after the intervention.

[0299] (5) Changes in exercise behavior: Record the change in the number of weeks with ≥3 days of exercise per week before and after the intervention ΔExercise.

[0300] (6) Changes in infant vital signs (applicable postpartum): Calculate the change in the number of days with abnormal infant vital signs (≥4 times / day) before and after the intervention.

[0301] It should be noted that only changes in the scale exceeding the minimum clinically significant difference (MCID) are considered clinically effective: EPDS change ≥ 3 points, PSQI change ≥ 2 points, and MSPSS change ≥ 5 points. Changes below the MCID are treated as zero contribution in subsequent validity coefficient calculations to exclude interference from scale measurement errors and minor random fluctuations.

[0302] 3. Calculation of effectiveness coefficient

[0303] The actual effectiveness coefficient E of each multimodal composite intervention scheme for a specific user is calculated. The theoretical range of the value is [-1, 1]. After formula correction, a symmetrical interval is achieved.

[0304] The formula for calculating E is as follows:

[0305] E = 0.35 × EPDS Improvement Rate + 0.15 × PSQI Improvement Rate + 0.10 × MSPSS Improvement Rate + 0.15 × Completion Rate + 0.15 × Life Behavior Score + 0.10 × Infant Safety Score.

[0306] The definitions and calculation methods for the preset weights of EPDS improvement rate, PSQI improvement rate, MSPSS improvement rate, CompletionRate standardization, LifeBehaviorScore and InfantSafetyScore are as follows.

[0307] (1) EPDS improvement rate, with values ​​[-1, 1].

[0308] EPDS improvement rate = (baseline EPDS - retested EPDS) / baseline EPDS.

[0309] When the retested EPDS is lower than the baseline, the improvement rate is positive (improved mood). When the retested EPDS is higher than the baseline, the improvement rate is negative (worsened mood).

[0310] (2) PSQI improvement rate, with values ​​of [-1, 1].

[0311] PSQI improvement rate = (baseline PSQI - retested PSQI) / baseline PSQI.

[0312] When the retested PSQI is lower than the baseline, the improvement rate is positive (sleep improvement). When the retested PSQI is higher than the baseline, the improvement rate is negative (sleep deterioration).

[0313] (3) MSPSS improvement rate, with values ​​of [-1, 1].

[0314] MSPSS improvement rate = (retested MSPSS - baseline MSPSS) / (84 - baseline MSPSS);

[0315] When the retested MSPSS is higher than the baseline, the improvement rate is positive (increased social support). When the retested MSPSS is lower than the baseline, the improvement rate is negative (deteriorated social support). The denominator is corrected using (84 - baseline MSPSS) to ensure comparability of improvement rates across different baseline levels.

[0316] (4) The CompletionRate is standardized and takes the value [-1, 1].

[0317] Linearly map the original completion rate (range [0, 1]) to the interval [-1, 1]:

[0318] CompletionRate (standardized) = 2 × original completion rate - 1.

[0319] For example, a 100% completion rate is mapped to +1, a 0% completion rate to -1, and a 50% completion rate to 0. This mapping ensures that the completion rate dimension has the same symmetrical interval as the scale's improvement rate, without lowering the lower limit of E.

[0320] (5) LifeBehaviorScore, with values ​​of [-1, 1].

[0321] Improvement in both dietary and exercise dimensions of lifestyle behavior:

[0322] LifeBehaviorScore = (ΔDiet + ΔExercise) / 2;

[0323] Wherein, ΔDiet is the change in the rate of calorie intake compliance (the percentage of days falling within 80%-120% of the recommended intake) before and after the intervention, with a value of [-1, 1]; ΔExercise is the change in the number of weeks with ≥3 days of exercise per week before and after the intervention, divided by the maximum possible change value, with a value of [-1, 1].

[0324] (6) Infant Safety Score, with values ​​of [-1, 1], applicable postpartum.

[0325] Based on the change in the number of days with abnormal vital signs in infants (≥4 times / day):

[0326] InfantSafetyScore = (Number of abnormal days before intervention - Number of abnormal days after intervention) / Number of abnormal days before intervention;

[0327] When the number of abnormal days after intervention is lower than before intervention, the InfantSafetyScore is positive (infant vital signs improve, maternal anxiety sources decrease); when the number of abnormal days after intervention is higher than before intervention, the InfantSafetyScore is negative (infant vital signs worsen, maternal anxiety sources increase). Since there is no data on infant vital signs in the pregnancy pattern, this dimension's weight is proportionally allocated to the EPDS improvement rate.

[0328] 4. Dynamic strategy optimization

[0329] The system performs a strategy optimization task every 30 days, marking intervention content with an actual effectiveness coefficient E below a preset threshold as inefficient content and reducing its matching weight; and marking content with an actual effectiveness coefficient E above the preset threshold as efficient content and increasing its matching weight. The actual effectiveness coefficient E of each intervention content is recalculated periodically to achieve continuous iterative optimization of the intervention strategy.

[0330] (1) Calculate the mean of the actual effectiveness coefficient of each intervention content in the most recent 30 days, mark the intervention content with a mean < 0 as "invalid content", and multiply the matching weight by 0.4.

[0331] (2) Intervention content with a mean of <0.30 is marked as “inefficient content” and the matching weight is multiplied by 0.7.

[0332] (3) Intervention content with a mean > 0.70 is marked as “high-efficiency content” and the matching weight is multiplied by 1.5.

[0333] (4) For newly added intervention content (content used for the first time in the past 30 days), the initial weight is set to 1.0, and it will be included in the optimization evaluation after accumulating at least 5 effectiveness score data.

[0334] (5) The basic matching weight of all intervention content is 1.0, and the weight range after optimization and adjustment is limited to [0.2, 3.0].

[0335] 5. Regular calibration

[0336] The system performs a weight calibration task every 180 days (6 months), as detailed below:

[0337] (1) Statistically analyze the distribution of the actual effectiveness coefficient E of all intervention programs in the past 180 days, and calculate the 33rd percentile and 67th percentile of E.

[0338] (2) The 33rd and 67th percentiles of E are used as the new dividing thresholds between “efficient” and “inefficient” to replace the fixed thresholds (i.e. 0.30 and 0.70) in the dynamic optimization of the strategy, so as to achieve adaptive adjustment of the judgment threshold.

[0339] The working principle of the above technical solution is as follows: user interaction behavior is recorded through process feedback, the actual clinical effect of the intervention is quantified through effect feedback, the effectiveness coefficient is used to evaluate both factors, and the strategy is adaptively adjusted through monthly regular optimization.

[0340] The effects of the above technical solution are as follows: process feedback identifies content that users are not interested in and downgrades its weight, thereby improving the acceptance rate of subsequent push notifications; effect feedback quantifies the actual effect of intervention, avoiding "push notifications and then ignoring them"; the effectiveness coefficient integrates process and effect to ensure the comprehensiveness of optimization decisions; and the periodic calibration function enables the judgment criteria to be dynamically optimized with system operation data, maintaining the continuous rationality of the evaluation system.

[0341] Example 3: Complete System Operation Flow Based on Mini Programs

[0342] In this embodiment, the user completes the following operation process through the client of a certain software program:

[0343] Step 1: User registration and stage identification.

[0344] When a user first enters the software, the system guides them to complete basic information, including: name, age, due date (or delivery date), and current gestational week (or postpartum days). The system automatically determines the user's stage based on the due date—if the current date is before the due date, it is marked as being in "pregnancy mode"; if the current date is after the due date but no more than 365 days later, it is marked as being in "postpartum mode"; if the current date is more than 365 days after the due date, the system prompts the user that they are outside the perinatal monitoring range and the monitoring function is disabled.

[0345] Step 2: Pregnancy / Postpartum Mode Operation Procedures.

[0346] In pregnancy mode, the software interface displays five function entry points:

[0347] (1) Emotional assessment entry: After clicking, the user enters the EPDS scale page, completes the scoring of 10 items, and submits the score. The system will complete the score calculation within 1 second and display the current risk level. If it is high risk, a red warning banner will be displayed at the top of the page.

[0348] (2) Sleep assessment entry: After clicking, users will enter the PSQI scale page and complete it once a month.

[0349] (3) Social support assessment entry: After clicking, users will enter the MSPSS scale page and complete it once a month.

[0350] (4) Dietary record entry: Users can click to enter the dietary record page every day, select the breakfast / lunch / dinner / snack time, add food names and portions (grams), and the system will calculate the calories and macronutrients of the meal in real time based on the preset food database.

[0351] (5) Exercise record entry: Users can click to enter the exercise record page every day, select the type of exercise, fill in the duration of exercise (minutes), and select the subjective exercise intensity (0-10 level slider).

[0352] Postpartum Mode Operation Procedure. In postpartum mode, the software program retains the five functional entry points mentioned above and adds a sixth functional entry point:

[0353] (6) Infant Monitoring Entry: After clicking, the user enters the infant vital signs page. The top of the page displays the currently bound radar device ID and online status. The middle of the page displays a list of the latest 10 vital signs data. Users can manually enter the infant's heart rate and respiratory data through the bottom button. If the frequency of abnormal vital signs is ≥4 times / day, a yellow warning banner will be displayed at the top of the page.

[0354] Step 3: Intervene in the content reception process.

[0355] Once the system triggers an alert or completes a periodic risk assessment, a personalized intervention plan is generated. Users can view the list of pushed intervention content in the "Messages" section on the software's homepage. Clicking on any content will take you to the details page, where the content for each modality is displayed in the order of primary intervention targets and secondary intervention targets.

[0356] (1) Cognitive behavioral guidance content: presented in the form of pictures, text and audio, including guiding text, practice steps and audio player. The system records whether the user scrolls to the bottom and the audio playback progress.

[0357] (2) Relaxing audiovisual content: presented in a video player, lasting 5-10 minutes. The system records the playback progress.

[0358] (3) Lifestyle adjustment suggestions: presented in the form of graphic cards, with a "Confirm Read" button at the bottom of the card.

[0359] (4) Mother-infant interaction guidance (postpartum): presented in a combination of video and text, including operation demonstration videos and a text list of key operation points.

[0360] Step 4: Data Synchronization and System Processing Flow.

[0361] All data submitted by the user on the software side is uploaded to the server in real time via HTTPS protocol. Upon receiving the data, the server performs the following processing (processing time ≤ 2 seconds):

[0362] (1) Data format validation: Validate whether the required fields are complete and whether the numerical range is valid. Return an error message for invalid data;

[0363] (2) Data storage: Data that passes verification is written to the MySQL database, and data with high real-time requirements is written to the Redis cache;

[0364] (3) Feature extraction trigger: A scheduled task at 2:00 AM every day extracts the previous day's data and updates the user's temporal feature vector;

[0365] (4) Association graph update: An association graph update is triggered every 14 days of newly added valid data;

[0366] (5) Risk prediction operation: The risk assessment is run immediately after the user submits the EPDS assessment, and an assessment is run once a day at 2:00 am for all active users;

[0367] (6) Solution generation and push: If an early warning is triggered after the risk assessment is completed, a personalized solution will be generated within 1 minute and pushed to the user terminal in real time through the message interface of the software program.

[0368] The working principle of the above technical solution is as follows: the system realizes the full-process digitalization and automation of perinatal emotional health management. Users complete operations such as registration, stage identification, data entry of various dimensions, and reception of intervention content through software programs. The server completes automated processing of data verification, storage, feature extraction, correlation map updates, risk prediction, and solution push.

[0369] The effects of the above technical solution are as follows: the system automatically identifies the pregnancy / postpartum pattern based on the expected delivery date; data collection entry points for each dimension are dynamically displayed in stages; fixed parameters at each stage on the server side ensure system stability; software program message push ensures timely intervention; and end-to-end automation reduces the human and time costs of perinatal emotional health management.

[0370] The above detailed embodiments describe the implementation of the present invention; however, the present invention is not limited to the specific details described in the above embodiments. Within the scope of the claims and technical concept of the present invention, various simple modifications and changes can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

Claims

1. A method for monitoring and managing perinatal emotional health based on multidimensional data correlation, characterized in that, Includes the following steps: Collect maternal pregnancy and postpartum data and infant vital signs data, perform preprocessing and temporal feature extraction, and construct corresponding temporal feature vectors; Align the temporal feature vectors of the pregnancy and postpartum data and the infant vital signs data along the time axis, establish the association edges between the pregnancy and postpartum data and the infant vital signs data, and construct a data association map of mother and infant; Based on the data association map, the association between emotional state data in the pregnancy and postpartum data and other pregnancy and postpartum data and infant vital signs data is identified, and a multidimensional association rule set is generated. Based on real-time collected pregnancy and postpartum data and infant vital signs data, as well as the multidimensional association rule set, the perinatal impact risk assessment of the mother is carried out, and the emotional risk assessment results of the mother are obtained. A multimodal intervention content library is constructed. Based on real-time collected pregnancy and postpartum data, infant vital signs data, and emotional risk assessment results, the current problem dimensions and the hit results in the multidimensional association rule set are identified. Personalized multimodal composite intervention plans are generated by matching the multimodal intervention content library. Track the mother's implementation of the multimodal intervention program and her emotional changes to dynamically optimize subsequent intervention strategies.

2. The perinatal emotional health monitoring and management method based on multidimensional data association according to claim 1, characterized in that, The pregnancy and postpartum data include emotional state data, sleep quality data, social support data, dietary and nutritional data, and exercise data during and after pregnancy and postpartum; the infant vital signs data include the infant's heart rate and respiratory rate. Furthermore, the emotional state data was collected using the Perinatal Emotional Self-Assessment Scale; the sleep quality data was collected using the Sleep Quality Assessment Scale; the social support data was collected using the Perinatal Support Capacity Assessment Scale; the dietary nutrition data and exercise data were obtained through manual input; and the infant vital signs data were obtained through both manual input and automatic collection.

3. The perinatal emotional health monitoring and management method based on multidimensional data association according to claim 1, characterized in that, The preprocessing method is as follows: imputing missing values, detecting and correcting outliers in the pregnancy and postpartum data and infant vital signs data; and normalizing the pregnancy and postpartum data and infant vital signs data according to their respective dimensions.

4. The perinatal emotional health monitoring and management method based on multidimensional data association according to claim 1, characterized in that, The associated edges include time-lag associated edges, same-time associated edges, and trend-related associated edges; Furthermore, the time lag correlation edge includes: taking the mother's emotional state data at the current moment as the dependent variable and other pregnancy and postpartum data and infant vital signs data from the past preset time as independent variables, calculating the correlation coefficient between the other pregnancy and postpartum data and infant vital signs data from the past preset time and the emotional state data, and determining the optimal lag time window for the other pregnancy and postpartum data and infant vital signs data to affect the emotional state data; The simultaneous correlation edge includes: within the same time window, calculating the correlation coefficient between the emotional state data and other pregnancy and postpartum data, and infant vital signs data; The trend correlation edge includes: calculating the correlation coefficient between the changes in the pregnancy and postpartum data and the infant vital signs data within a preset time window.

5. The perinatal emotional health monitoring and management method based on multidimensional data association according to claim 1, characterized in that, Based on the data association map, the method for identifying the associations between emotional state data in the pregnancy and postpartum data and other pregnancy and postpartum data, as well as infant vital sign data, and generating a multidimensional association rule set is as follows: The feature values ​​of pregnancy and postpartum data and infant vital signs data are discretized into different level labels according to preset thresholds to form a level label sequence of each dimension. The FP-Growth algorithm is used to perform frequent pattern mining on the level label sequence to obtain several association rules. The association rules are then filtered and sorted according to support and confidence to generate a multidimensional association rule set.

6. The perinatal emotional health monitoring and management method based on multidimensional data association according to claim 1, characterized in that, Based on real-time collected pregnancy and postpartum data and infant vital signs data, as well as the aforementioned multidimensional association rule set, methods for assessing the perinatal impact risks of mothers and obtaining the results of the mother's emotional risk assessment include: Using the current pregnancy and postpartum data and infant vital signs data, as well as the association rule matching features extracted from the multidimensional association rule set as the evaluation basis, the user's current emotional risk assessment result is determined and output. The emotional risk assessment results include the emotional risk level and the emotional risk trend within a preset future time window.

7. The perinatal emotional health monitoring and management method based on multidimensional data association according to claim 1, characterized in that, The multimodal intervention content library includes cognitive behavior guidance content, soothing audiovisual content, lifestyle and support system adjustment suggestions, and maternal-infant interaction guidance. Furthermore, the lifestyle and support system adjustment recommendations include sleep hygiene education content, family communication promotion guidelines, social support activation recommendations, personalized nutrition recommendations, and personalized exercise programs; Furthermore, the mother-infant interaction guidance includes suggestions for parent-child interaction and guidance on soothing techniques based on infant vital signs data.

8. The perinatal emotional health monitoring and management method based on multidimensional data association according to claim 1, characterized in that, The method for tracking the mother's implementation of the multimodal intervention program and changes in her emotions, and dynamically optimizing subsequent intervention strategies, is as follows: Record the mother's viewing, completion, and execution of various intervention components in the multimodal composite intervention plan as process feedback; Collect pregnancy and postpartum data and infant vital signs data after the intervention of the multimodal composite intervention program as feedback on the effect; Set coefficient thresholds; Based on the process feedback and effect feedback, calculate the actual effectiveness coefficient of each type of intervention content in the multimodal composite intervention program; Intervention content with an actual effectiveness coefficient lower than a preset threshold is marked as inefficient content, and its matching weight is reduced. Intervention content with an actual effectiveness coefficient higher than a preset threshold is marked as high-efficiency content and its matching weight is increased.

9. A perinatal emotional health monitoring and management system based on multidimensional data association, characterized in that, Used for performing the perinatal emotional health monitoring and management method as described in any one of claims 1-8; The perinatal emotional health monitoring and management system includes: The multidimensional data acquisition and processing module is used to collect data on the mother's pregnancy and postpartum period. The IoT data access module is used to receive infant vital sign data collected by IoT radar sensors and perform empty room filtering and binding verification. The temporal feature extraction module is used to preprocess and extract temporal features from the pregnancy and postpartum data and infant vital signs data, and construct the corresponding temporal feature vector. The data association graph construction module is used to align the time-series feature vectors of the pregnancy and postpartum data and the infant vital signs data along the time axis, establish the pregnancy and postpartum data and the infant vital signs data, and construct a data association graph between the mother and the infant. The association rule mining module is used to identify the association between emotional state data in the pregnancy and postpartum data and other pregnancy and postpartum data and infant vital signs data based on the data association graph, and generate a multidimensional association rule set. The perinatal emotional risk assessment module is used to assess the perinatal impact risk of mothers based on real-time collected pregnancy and postpartum data and infant vital signs data, as well as the multidimensional association rule set, and to obtain the emotional risk assessment results of mothers. The personalized intervention module is used to build a multimodal intervention content library. Based on real-time collected pregnancy and postpartum data, infant vital signs data, and emotional risk assessment results, it identifies the current problem dimensions and the hit results in the multidimensional association rule set, and generates personalized multimodal composite intervention plans from the multimodal intervention content library. It tracks the mother's implementation of the multimodal composite intervention plan and emotional changes, and dynamically optimizes subsequent intervention strategies.

Images