Multi-modal sensing system for dynamic evaluation of labor process

Abstract

The invention discloses a multi-modal sensing system for dynamic evaluation of a labor process. The multi-modal sensing system comprises a signal acquisition module, a uterine contraction pressure acquisition module, a pregnant woman clinical information acquisition module, a data processing module and an intelligent decision output module, the signal acquisition module is used for acquiring fetal signals in real time; the uterine contraction pressure acquisition module is used for acquiring a uterine contraction intensity signal of a pregnant woman; the pregnant woman clinical information acquisition module is used for acquiring clinical information data of a pregnant woman; the data processing module is used for generating fetal orientation evaluation information according to the fetal signal, obtaining parturition progress information according to the uterine contraction intensity signal and generating a dynamic risk score according to the clinical information of the pregnant woman; and the intelligent decision output module generates delivery risk early warning information according to the fetal orientation evaluation information, the delivery progress information and / or the dynamic risk score. The birth process evaluation objectivity and accuracy can be improved, and a scientific basis is provided for birth management.

Description

Technical Field

[0001] The present invention belongs to the field of medical technology, and specifically relates to a multimodal sensing system for dynamic assessment of labor process. Background Art

[0002] In modern obstetric clinical practice, accurate and dynamic assessment of labor progress is crucial for ensuring maternal and infant safety and optimizing delivery management. However, current primary methods for assessing labor progress and determining key information such as fetal position have significant shortcomings. For example, the traditional digital vaginal examination relies heavily on subjective judgment by the examiner and lacks precision. Studies have shown that its error rate can reach as high as 20%-70%. Furthermore, this invasive procedure not only increases maternal discomfort but also increases the risk of intrauterine infection. Meanwhile, cardiotocography (CTG), a key form of electronic fetal monitoring, records fetal heart rate changes and their temporal relationship with uterine contraction pressures, making it crucial for assessing fetal condition. However, CTG interpretation relies heavily on the obstetrician's experience and is subject to significant subjective factors. This is particularly true in primary care hospitals in my country, where the risk of misdiagnosis is high due to physician inexperience.

[0003] In recent years, artificial intelligence (AI) technology has been increasingly applied to intelligent CTG interpretation, but existing methods still have numerous limitations. Traditional machine learning methods rely on feature extraction from CTG signals, but the subjectivity of manually formulated extraction rules is prominent, and secondary feature extraction is prone to introducing errors. While deep learning-based methods can directly extract deep features from fetal heart rate (FHR) signals, most methods overlook the importance of uterine contraction pressure signals (UC) and maternal clinical information (such as gestational age and maternal age), failing to fully exploit the rich information in multimodal data. Meanwhile, intrapartum ultrasound technology, due to its accuracy and reproducibility, has demonstrated significant value in labor assessment. However, this technology requires a high level of operator skill, and the long learning curve and lack of practical experience have limited its widespread application. Traditional monitoring methods are generally subject to lags and subjectivity, making it difficult to comprehensively and real-timely assess fetal status. Summary of the Invention

[0004] The purpose of the present invention is to provide a multimodal sensing system for dynamic assessment of labor process. The present invention can improve the objectivity and accuracy of labor process assessment and provide a scientific basis for delivery management.

[0005] The technical solution of the present invention is a multimodal sensing system for dynamic assessment of labor process, comprising a signal acquisition module, a uterine contraction pressure acquisition module, a pregnant woman's clinical information acquisition module, a data processing module and an intelligent decision output module; The signal acquisition module is used to obtain fetal signals in real time; The uterine contraction pressure acquisition module is used to collect the uterine contraction intensity signal of the pregnant woman; The clinical information acquisition module for pregnant women is used to obtain clinical information data of pregnant women; The data processing module generates fetal position assessment information based on fetal signals, obtains labor progress information based on uterine contraction intensity signals, and generates a dynamic risk score based on the clinical information of the pregnant woman; The intelligent decision output module generates delivery risk warning information based on fetal position assessment information, labor progress information and / or dynamic risk score.

[0006] In the above-mentioned multimodal sensing system for dynamic assessment of the labor process, the acquisition equipment of the signal acquisition module includes a fetal heart monitoring sensor, a micro-acceleration sensor and a temperature sensor; the signal acquisition module monitors the fetal heart rate signal in real time through the fetal heart monitoring sensor; the signal acquisition module captures the fetal movement signal caused by fetal movement through the micro-acceleration sensor; the signal acquisition module detects the temperature changes of the contact points between the fetal heart monitoring sensor, the micro-acceleration sensor and the pregnant woman's skin in real time through the temperature sensor to evaluate the signal acquisition quality.

[0007] In the aforementioned multimodal sensing system for dynamic assessment of labor progress, the fetal position assessment information generated by the data processing module includes fetal heart signal spectrum deviation and fetal movement trajectory; The fetal heart signal spectrum offset is obtained by extracting the main frequency component of the fetal heart beat frequency from the fetal heart rate signal and comparing it with the standard fetal heart frequency range to calculate the spectrum offset and its change trend; The fetal motion trajectory is reconstructed based on the fetal motion signal captured by the micro-acceleration sensor, and the position deviation of the fetal head relative to the maternal pelvis is calculated; According to the formula Determine whether the fetal position is abnormal, including: is the spectrum offset, Deviation in the position of the fetal head.

[0008] In the aforementioned multimodal sensing system for dynamic assessment of labor progress, the specific process of the uterine contraction pressure signal acquisition module acquiring the uterine contraction intensity signal is as follows: A flexible pressure distribution sensor array is placed on the abdomen of pregnant women to capture the changes in pressure distribution during uterine contractions in real time; Based on the pressure distribution sensor data, using the formula Quantify the contraction intensity signal, where is the quantitative value of the uterine contraction intensity signal, For the The pressure value of each sensor unit, is the effective area of ​​the sensor unit, is the total number of sensor units.

[0009] In the aforementioned multimodal sensing system for dynamic assessment of labor progress, the labor progress information generated by the data processing module includes labor stage division and fetal head descent assessment: The labor stage division is based on the dynamic changes of uterine contraction intensity signals and fetal signals, using a hidden Markov model to divide the labor stages and calculate the probability distribution of the current labor stage; The fetal head descent assessment is based on the fetal head movement trajectory data using the formula Assess the rate of descent of the fetal head, including: is the fetal head descent speed, is the change in fetal head position, is the time interval.

[0010] In the aforementioned multimodal perception system for dynamic assessment of labor progress, the process of the pregnant woman's clinical information acquisition module acquiring the pregnant woman's clinical information data is as follows: Real-time monitoring of pregnant women's blood pressure, heart rate, and blood oxygen saturation through wireless wearable devices and combined with electronic health records to obtain gestational age, age, and body mass index; The support vector machine algorithm is used to classify and evaluate the clinical information of pregnant women and calculate the dynamic risk score:

[0011] in, is a dynamic risk score, For blood pressure, is the heart rate, is the body mass index, 、 and is the weight coefficient, is the bias term.

[0012] In the aforementioned multimodal perception system for dynamic assessment of labor progress, the delivery risk warning information includes: When the position deviation Exceeding the preset threshold When the fetal position is abnormal, When the fetal head descent speed is lower than the threshold When the uterine contraction intensity signal is lower than the normal range, the labor stage is determined to be abnormal; When the dynamic risk score exceeds the preset threshold, the pregnant woman's delivery is judged to be abnormal.

[0013] In the aforementioned multimodal perception system for dynamic assessment of labor progress, the data processing module also includes an environmental noise suppression function, and the specific process is as follows: Extract noise features from collected fetal signals, uterine contraction intensity signals, and clinical information of pregnant women, and use variational autoencoders to model the noise; By formula Optimize the noise suppression model, where is the loss function, is the KL divergence, is the encoder distribution, is the prior distribution, is the decoder distribution; When the noise suppression effect reaches a preset standard, environmental noise suppression information is generated.

[0014] In the aforementioned multimodal sensing system for dynamic assessment of labor progress, the data processing module also includes a multi-source data correction function, the specific process of which is as follows: Extract the timestamp information of fetal signals, uterine contraction intensity signals, and maternal clinical information, align the time axes of the three, and calculate the time deviation ; ; Where, 、 and They are the timestamps of fetal heart rate signal, uterine contraction pressure signal and clinical information of pregnant women; When time deviation Generate multi-source data correction information when it exceeds a preset threshold.

[0015] The aforementioned multimodal perception system for dynamic assessment of labor process, wherein the intelligent decision output module sends delivery risk warning information to the obstetrician terminal device via a visual interface or wireless communication.

[0016] Compared to existing technologies, the present invention achieves simultaneous multi-source data collection through a signal acquisition module, a uterine contraction pressure acquisition module, and a maternal clinical information acquisition module, avoiding the assessment limitations of a single data dimension. The present invention dynamically acquires fetal position assessment information, labor progress information, and dynamic risk scores through a data processing module, avoiding the subjectivity inherent in relying on physician experience to interpret CTGs. This particularly reduces the risk of misdiagnosis due to physician inexperience in primary care hospitals. The present invention generates real-time labor risk warnings (such as abnormal fetal position, abnormal labor stage, and abnormal maternal labor) through an intelligent decision-making output module and pushes these warnings to physician terminals via a visual interface or wireless communication, facilitating timely intervention. This enables precise assessment of labor dynamics, overcoming the high subjectivity and low accuracy of existing methods. Furthermore, through deep fusion processing of multimodal data, it provides scientific decision support for obstetricians. Furthermore, the present invention uses a variational autoencoder to suppress environmental noise and uses timestamps to align multi-source data, ensuring data reliability and avoiding assessment lags caused by signal interference or time deviation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1It is a schematic diagram of the system structure of the present invention; Figure 2 This is a schematic diagram of the fetal signal acquisition and processing process; Figure 3 It is a schematic diagram of the process of processing labor progress information; Figure 4 It is a flowchart of the environmental noise suppression and multi-source data correction functions. DETAILED DESCRIPTION

[0018] The present invention will be further described below with reference to the accompanying drawings and examples, but they are not intended to limit the present invention.

[0019] Example: A multimodal sensing system for dynamic assessment of labor process, such as Figure 1 As shown, the system includes a signal acquisition module, a uterine contraction pressure acquisition module, a maternal clinical information acquisition module, a data processing module, and an intelligent decision-making output module. The signal acquisition module is used to acquire fetal signals in real time; the uterine contraction pressure acquisition module is used to collect the maternal contraction intensity signal; and the maternal clinical information acquisition module is used to obtain the maternal clinical information data. The data processing module generates fetal position assessment information based on fetal signals, obtains labor progress information based on uterine contraction intensity signals, and generates a dynamic risk score based on the maternal clinical information. The intelligent decision-making output module generates delivery risk warning information based on the fetal position assessment information, labor progress information, and dynamic risk score. The intelligent decision-making output module can transmit delivery risk warning information to the doctor's terminal via a visual interface or wireless communication.

[0020] Preferably, the signal acquisition module's acquisition equipment includes a fetal heart rate monitoring sensor, a micro-accelerometer, and a temperature sensor. The signal acquisition module uses the fetal heart rate monitoring sensor to monitor fetal heart rate signals in real time. The signal acquisition module uses the micro-accelerometer to capture fetal movement signals. The signal acquisition module uses the temperature sensor to monitor temperature changes at the contact points between the fetal heart rate monitoring sensor and the micro-accelerometer and the pregnant woman's skin in real time to assess signal acquisition quality. Specifically, the fetal heart rate monitoring sensor can be a Doppler ultrasound probe to collect fetal cardiac electrical activity signals. The Doppler ultrasound probe operates at a frequency of 2-5 MHz, enabling real-time fetal heart rate signals. The sampling frequency is set to 100 Hz to ensure the accuracy of heart rate data. The Doppler ultrasound probe is fixed to the pregnant woman's abdomen at the location where fetal heart sounds are strongest, using a coupling agent to ensure effective sound transmission. The micro-accelerometer is preferably a MEMS triaxial accelerometer (such as the ADXL345), with a range of 0.1-10g and a sampling frequency of at least 100 Hz. This allows for accurate capture of weak vibration signals caused by fetal movement. The micro-accelerometers are symmetrically arranged around the ultrasound probe to capture three-dimensional acceleration signals of fetal head movement. The temperature sensor uses a flexible thermocouple array with a measurement accuracy of ±0.1°C. The temperature sensor array surrounds the sensor attachment area and records temperature data in real time. When the temperature fluctuation exceeds ±0.5°C, it indicates that the sensor attachment state is abnormal. Therefore, by monitoring the temperature gradient change between the sensor and the skin contact surface, it is determined whether the sensor attachment state is stable. Furthermore, based on the data collected by the signal acquisition module, the fetal position assessment information generated by the data processing module includes the fetal heart signal spectrum deviation and fetal movement trajectory, such as Figure 2 As shown: The calculation process of the fetal heart signal spectrum offset includes: converting the fetal heart rate signal into a frequency domain signal through fast Fourier transform and extracting the main frequency component; comparing the main frequency component with the standard fetal heart rate frequency range (110-160bpm) to calculate the spectrum offset; and analyzing the changing trend of the spectrum offset through a sliding window (length of 5 minutes).

[0021] The fetal motion trajectory reconstructs the fetal head trajectory based on fetal motion signals captured by a micro-accelerometer and calculates the positional deviation of the fetal head relative to the maternal pelvis. Specifically, a micro-accelerometer (MEMS triaxial accelerometer) placed on the pregnant woman's abdomen captures triaxial acceleration signals (x, y, and z axes) caused by fetal movement at high frequency (e.g., 100 Hz), reflecting the vibration, rotation, and other movements of the fetal head within the mother's body. The acceleration signal is subjected to noise reduction (e.g., Kalman filtering) to eliminate environmental noise interference. The acceleration signal is then converted into three-dimensional displacement data through two discrete integrations (first acceleration → velocity, then velocity → displacement). Combined with the physical coordinates of the sensor on the pregnant woman's abdomen (e.g., establishing a three-dimensional coordinate system with the pubic symphysis as the origin), the displacement data at each moment is fitted into a continuous fetal head motion trajectory, forming a model of the fetal spatial motion path within the mother's body. Using anatomical landmarks of the maternal pelvis (e.g., the ischial spine and sacral promontory) as reference points, the three-dimensional spatial deviation between the real-time position of the fetal head in the trajectory and the ideal delivery position (e.g., occiput anterior position) is calculated. , according to the deviation, determine whether the fetal position is abnormal, among which, is the spectrum offset, Deviation in the position of the fetal head.

[0022] Preferably, the specific process of the uterine contraction pressure signal acquisition module acquiring the uterine contraction intensity signal is as follows: A flexible pressure distribution sensor array is arranged on the abdomen of the pregnant woman to capture the changes in pressure distribution during uterine contractions in real time. The flexible pressure distribution sensor array consists of multiple micro pressure sensor units arranged in a matrix. The sensor units are encapsulated with a polyimide substrate with a thickness of no more than 1 mm, which can fit the curved surface of the pregnant woman's abdomen. The effective area of ​​each sensor unit is 4 square millimeters, and the total coverage area of ​​the array is not less than 200 square centimeters. In this embodiment, a flexible array consisting of 128 micro pressure sensor units is attached to the area from the pubic symphysis to the fundus of the uterus in the pregnant woman's abdomen, with a unit spacing of 5 mm and a total coverage area of ​​20 cm × 15 cm. Each sensor unit adopts the piezoresistive principle, with an effective area of ​​4 mm² and a sampling frequency of 50 Hz. The pressure value is converted into an electrical signal through the piezoresistive effect, and the sampling frequency is set to 50 Hz to ensure dynamic capture of uterine contraction pressure fluctuations.

[0023] Based on the pressure distribution sensor data, using the formula Quantify the contraction intensity signal, where is the quantitative value of the uterine contraction intensity signal, For the The pressure value of each sensor unit, is the effective area of ​​the sensor unit, is the total number of sensor units.

[0024] In the calculation formula for the contraction intensity signal, the total number of sensor units is determined by the actual array size deployed, typically ranging from 100 to 200 units. This allows the spatial distribution of contraction pressure to be captured through a high-density flexible sensor array, resolving the problem of traditional single-point pressure sensors being unable to fully reflect the spatial differences in contraction intensity signals. This allows for a more accurate reflection of the true intensity of contractions, avoiding assessment bias caused by localized measurements. Furthermore, the real-time dynamic data acquisition method overcomes the subjectivity and intermittent nature of traditional manual palpation, providing objective and continuous data support for labor progress assessment.

[0025] Preferably, the labor progress information generated by the data processing module includes labor stage division and fetal head descent assessment: like Figure 3 As shown, the labor stage division is based on the dynamic changes of uterine contraction intensity signals and fetal signals, using a hidden Markov model to divide the labor stages and calculate the probability distribution of the current labor stage. The hidden Markov model probabilistically models the labor stages using an observation sequence consisting of uterine contraction intensity signals and fetal signals. The uterine contraction intensity signals are collected by an array of flexible pressure distribution sensors, and the fetal signals include fetal heart rate signals and fetal movement signals. The hidden states of the hidden Markov model correspond to different labor stages, including the latent phase, active phase, and second stage of labor. As a preferred embodiment, the hidden Markov model uses the Baum-Welch algorithm for parameter training, and the probability distribution of the current labor stage is calculated using a forward-backward algorithm. The specific process is as follows: First, the transition probability matrix A, the emission probability matrix B, and the initial state probability π are initialized based on historical labor data. The historical data includes multimodal data such as uterine contraction intensity and fetal heart rate from 500 normal delivery cases. Then, the forward and backward variables are iteratively calculated using the forward-backward algorithm, and the model parameters are re-estimated until convergence. In real-time labor assessment, a forward-backward algorithm is used to calculate the posterior probability distribution of the current labor stage. For example, when the probability of the active phase exceeds 80% and lasts for 15 minutes, it is determined to have entered the active phase.

[0026] The fetal head descent assessment is based on the fetal head movement trajectory data using the formula Assess the rate of descent of the fetal head, including: is the fetal head descent speed, is the change in fetal head position, is the time interval. During fetal head descent assessment, the change in fetal head position is reconstructed using fetal motion signals captured by a miniature accelerometer. In specific implementation, a Kalman filter is used to reduce noise in the motion signal to improve the accuracy of position calculation. The time interval is determined by the sensor sampling frequency; for example, at a sampling frequency of 10 Hz, the time interval is 0.1 seconds. The formula for calculating the fetal head descent velocity can be further optimized to account for the angle between the fetal head movement direction and the maternal pelvic axis.

[0027] Preferably, the process of the pregnant woman clinical information acquisition module acquiring the pregnant woman's clinical information data is as follows: Pregnant women's blood pressure, heart rate, and blood oxygen saturation are monitored in real time through wireless wearable devices, and combined with electronic health records to obtain gestational age, age, and body mass index. Wireless wearable devices can be smart bracelets or chest-mounted sensors, measuring blood pressure and blood oxygen saturation using photoplethysmography and heart rate using accelerometers. The electronic health record system obtains structured data, including gestational age, age, and body mass index from the most recent prenatal checkup, through a hospital information system interface.

[0028] The support vector machine algorithm is used to classify and evaluate the clinical information of pregnant women and calculate the dynamic risk score:

[0029] in, is a dynamic risk score, For blood pressure, is the heart rate, is the body mass index, 、 and are weight coefficients, set to 0.4, 0.3 and 0.2 respectively, is a bias term, set to 0.1. The input layer of the support vector machine algorithm receives standardized multi-dimensional clinical parameters, and the kernel function uses the radial basis function, and the optimal hyperplane is determined by cross-validation. In the calculation formula of the dynamic risk score, the weight coefficient is obtained by training historical delivery data, and the bias term is used to adjust the score baseline. The present invention realizes real-time continuous monitoring and standardized data acquisition of key physiological parameters through the collaborative work of wireless wearable devices and electronic health record systems. The support vector machine algorithm can effectively handle nonlinear classification problems, and compared with the traditional logistic regression method, it has better modeling capabilities for complex interactions between clinical characteristics. The evaluation results, together with fetal position information and labor progress information, constitute a multidimensional early warning system, providing more comprehensive data support for clinical decision-making.

[0030] Preferably, the delivery risk warning information includes: When the position deviation Exceeding the preset threshold When the fetal position is abnormal, When the fetal head descent speed is lower than the threshold When the uterine contraction intensity signal is lower than the normal range, the labor stage is determined to be abnormal; When the dynamic risk score exceeds the preset threshold, the pregnant woman is judged to have an abnormal delivery. Specifically, the preset threshold of position deviation is determined by clinical data statistics. For example, an early warning is triggered when the position deviation of the fetal head relative to the maternal pelvis exceeds 30mm. The threshold of the fetal head descent speed is dynamically adjusted according to the stage of labor. For example, it is judged to be abnormal when it is less than 1cm / h in the active phase of the first stage of labor or less than 2cm / h in the second stage of labor. The normal range of uterine contraction intensity signals is defined as 3-5 uterine contractions per 10 minutes, and the single uterine contraction intensity signal reaches 30-50mmHg. The preset threshold of the dynamic risk score is determined by ROC curve analysis. For example, when the score exceeds 0.7, it is judged to be high risk. These judgment conditions are applied in combination through logical and / or relationships. For example, when the abnormal fetal position and abnormal labor stage are met at the same time, an emergency intervention warning is triggered. Thus, the problem of insufficient reliability of single indicator warning in traditional labor assessment is solved by the joint judgment of multiple parameters. This method replaces subjective judgment with quantitative indicators, avoiding the errors associated with digital vaginal examinations. It also integrates fetal physiological signals, uterine contraction parameters, and maternal clinical data, overcoming the limitations of single CTG signal analysis. Finally, a dynamic threshold mechanism adapts to the changing characteristics of different labor stages, offering greater clinical applicability than fixed thresholds. This method reduces the false alarm rate while improving the sensitivity of identifying critical situations.

[0031] Preferably, the data processing module also includes an environmental noise suppression function, such as Figure 4 As shown, the process is as follows: Extract the noise features from the collected fetal signals, uterine contraction intensity signals and clinical information of pregnant women, and use the variational autoencoder to model the noise. Among them, the noise feature extraction can be based on frequency domain analysis or time domain statistical methods, such as using wavelet transform to separate the signal and noise components. The encoder of the variational autoencoder can use a convolutional neural network structure, and the decoder can use a deconvolutional network structure. Then, through the formula Optimize the noise suppression model, where is the loss function, is the KL divergence, is the encoder distribution, is the prior distribution, Decoder distribution.

[0032] In KL divergence calculations, the prior distribution is typically set to a standard normal distribution. The effectiveness of noise suppression can be quantified using either the signal-to-noise ratio improvement or the signal distortion metric. The predefined criteria can be set as a signal-to-noise ratio improvement of ≥10dB and a signal distortion of ≤5%. After model training is complete, when a noisy signal is input, the encoder separates the noise features, and the decoder generates the noise-reduced signal. When the noise reduction effect meets the predefined criteria (e.g., a signal-to-noise ratio improvement of ≥10dB), the system generates a "surrounding noise suppression complete" message, providing a clean data foundation for subsequent process evaluation. Thus, by establishing a noise suppression model, the useful signal can be effectively separated from the ambient noise. Compared to traditional filtering methods, variational autoencoders can learn the deep statistical characteristics of noise, avoiding the limitations of manually set filtering parameters. By jointly optimizing the reconstruction error and KL divergence, noise suppression is achieved while maintaining signal fidelity.

[0033] Preferably, the data processing module also includes a multi-source data correction function, the specific process of which is as follows: Extract timestamp information of fetal signals, uterine contraction intensity signals, and maternal clinical information. These timestamp data are standardized through a time synchronization protocol, such as using the Network Time Protocol (NTP) or Precision Time Protocol (PTP) for clock synchronization, and then calculate the time deviation. ; ; Where, 、 and They are the timestamps of fetal heart rate signal, uterine contraction pressure signal and clinical information of pregnant women; When time deviation When the delay exceeds a preset threshold (e.g., 500 milliseconds), the system automatically triggers a data correction process, compensating the time axis of the delayed signal using linear or spline interpolation methods and generating multi-source data correction information containing correction parameters. This solves the problem of data time asynchrony caused by differences in sensor sampling frequencies or transmission delays in multimodal sensing systems. Thus, through precise timestamp alignment and deviation detection mechanisms, the present invention effectively avoids problems such as misjudgment of the fetal heart rate-uterine pressure relationship and incorrect correlation of clinical information with physiological signals caused by time asynchrony.

[0034] Preferably, the intelligent decision-making output module sends the delivery risk warning information to the obstetrician's terminal device through a visual interface or wireless communication. Specifically, the visual interface can be implemented in the following ways: deploy a graphical user interface on the delivery room monitoring terminal to dynamically display the changing trends of fetal heart rate, uterine contraction intensity signal and dynamic risk score in the form of a line graph, and distinguish different warning levels by color coding. Among them, the abnormal fetal position warning mark is yellow, the abnormal labor stage mark is orange, and the delivery abnormality mark is red. The wireless communication method can be implemented in the following ways: docking with the hospital information system through Bluetooth or Wi-Fi protocol, and automatically triggering a text message push to the doctor's mobile terminal when the warning information is generated. The message content includes the pregnant woman's ID, warning type and occurrence timestamp. Furthermore, the warning information transmission process uses the AES-256 encryption algorithm to ensure data security.

[0035] In summary, the present invention obtains fetal signals, uterine contraction intensity signals and clinical information of pregnant women in real time, uses a data processing module to generate fetal position assessment, labor progress and delivery risk warning information, and sends it to the doctor terminal through a visual interface or wireless communication, thereby improving the objectivity and accuracy of labor assessment and providing a scientific basis for delivery management.

Claims

1. A multimodal sensing system for dynamic assessment of labor progress, characterized by: It includes signal acquisition module, uterine contraction pressure acquisition module, pregnant women's clinical information acquisition module, data processing module and intelligent decision output module; The signal acquisition module is used to obtain fetal signals in real time; The uterine contraction pressure acquisition module is used to collect the uterine contraction intensity signal of the pregnant woman; The clinical information acquisition module for pregnant women is used to obtain clinical information data of pregnant women; The data processing module generates fetal position assessment information based on fetal signals, obtains labor progress information based on uterine contraction intensity signals, and generates a dynamic risk score based on the clinical information of the pregnant woman; The intelligent decision output module generates delivery risk warning information based on fetal position assessment information, labor progress information and / or dynamic risk score.

2. The multimodal sensing system for dynamic assessment of labor process according to claim 1, characterized in that: The acquisition equipment of the signal acquisition module includes a fetal heart monitoring sensor, a micro acceleration sensor and a temperature sensor; the signal acquisition module monitors the fetal heart rate signal in real time through the fetal heart monitoring sensor; the signal acquisition module captures the fetal movement signal caused by fetal movement through the micro acceleration sensor; the signal acquisition module detects the temperature changes of the contact points between the fetal heart monitoring sensor, the micro acceleration sensor and the pregnant woman's skin in real time through the temperature sensor to evaluate the signal acquisition quality.

3. The multimodal sensing system for dynamic assessment of labor process according to claim 2, characterized in that: The fetal position assessment information generated by the data processing module includes fetal heart signal spectrum deviation and fetal movement trajectory; The fetal heart signal spectrum offset is obtained by extracting the main frequency component of the fetal heart beat frequency from the fetal heart rate signal and comparing it with the standard fetal heart frequency range to calculate the spectrum offset and its change trend; The fetal motion trajectory is reconstructed based on the fetal motion signal captured by the micro-acceleration sensor, and the position deviation of the fetal head relative to the maternal pelvis is calculated; According to the formula Determine whether the fetal position is abnormal, including: is the spectrum offset, Deviation in the position of the fetal head.

4. The multimodal sensing system for dynamic assessment of labor process according to claim 3, characterized in that: The specific process of the uterine contraction pressure signal acquisition module acquiring the uterine contraction intensity signal is as follows: A flexible pressure distribution sensor array is placed on the abdomen of pregnant women to capture the changes in pressure distribution during uterine contractions in real time; Based on the pressure distribution sensor data, using the formula Quantify the contraction intensity signal, where is the quantitative value of the uterine contraction intensity signal, For the The pressure value of each sensor unit, is the effective area of ​​the sensor unit, is the total number of sensor units.

5. The multimodal sensing system for dynamic assessment of labor process according to claim 4, characterized in that: The labor progress information generated by the data processing module includes labor stage division and fetal head descent assessment: The labor stage division is based on the dynamic changes of uterine contraction intensity signals and fetal signals, using a hidden Markov model to divide the labor stages and calculate the probability distribution of the current labor stage; The fetal head descent assessment is based on the fetal head movement trajectory data using the formula Assess the rate of descent of the fetal head, including: is the fetal head descent speed, is the change in fetal head position, is the time interval.

6. The multimodal sensing system for dynamic assessment of labor process according to claim 5, characterized in that: The process of the clinical information acquisition module for pregnant women obtaining clinical information data of pregnant women is as follows: Real-time monitoring of pregnant women's blood pressure, heart rate, and blood oxygen saturation through wireless wearable devices and combined with electronic health records to obtain gestational age, age, and body mass index; The support vector machine algorithm is used to classify and evaluate the clinical information of pregnant women and calculate the dynamic risk score: in, is a dynamic risk score, For blood pressure, is the heart rate, is the body mass index, 、 and is the weight coefficient, is the bias term.

7. The multimodal sensing system for dynamic assessment of labor process according to claim 6, characterized in that: The delivery risk warning information includes: When the position deviation Exceeding the preset threshold When the fetal position is abnormal, When the fetal head descent speed is lower than the threshold When the uterine contraction intensity signal is lower than the normal range, the labor stage is determined to be abnormal; When the dynamic risk score exceeds the preset threshold, the pregnant woman's delivery is judged to be abnormal.

8. The multimodal sensing system for dynamic assessment of labor process according to claim 1, characterized in that: The data processing module also includes an environmental noise suppression function, and the specific process is as follows: Extract noise features from collected fetal signals, uterine contraction intensity signals, and clinical information of pregnant women, and use variational autoencoders to model the noise; By formula Optimize the noise suppression model, where is the loss function, is the KL divergence, is the encoder distribution, is the prior distribution, is the decoder distribution; When the noise suppression effect reaches a preset standard, environmental noise suppression information is generated.

9. The multimodal sensing system for dynamic assessment of labor process according to claim 8, characterized in that: The data processing module also includes a multi-source data correction function, the specific process is as follows: Extract the timestamp information of fetal signals, uterine contraction intensity signals, and maternal clinical information, align the time axes of the three, and calculate the time deviation ; ； Where, 、 and They are the timestamps of fetal heart rate signal, uterine contraction pressure signal and clinical information of pregnant women; When time deviation Generate multi-source data correction information when it exceeds a preset threshold.

10. The multimodal sensing system for dynamic assessment of labor process according to claim 1, characterized in that: The intelligent decision output module sends the delivery risk warning information to the obstetrician terminal device through a visual interface or wireless communication.

Images