Little Beats: Fetal Monitoring Belt

Abstract

The use of third-trimester fetal monitoring is very crucial in the detection of fetal distress, hypoxia, abnormal wall motion, and irregular fetal heart rate (FHR). The traditional techniques, such as Doppler ultrasound and cardiotocography (CTG), are mostly hospital-limited, need trained staff, and cannot be used either in continuous or home monitoring, especially in resource-restricted environments. This paper reports the design and development of a non-invasive wearable fetal monitoring system, which is able to continuously track the fetal movement (FM) and fetal heart rate (FHR) outside the clinical settings. The suggested solution is deployed in the form of a wearable belt that comprises sensitive sensors such as ADXL345 accelerometers that capture fetal movements (FM) and Ag/AgCl ECG electrodes that capture fetal heart rate (FHR) signals. Other components such as ADS1115 analog-todigital converters (ADCs) are used to condition the signal, and an ESP32 microcontroller is used to ensure real-time data acquisition and transmission. The device has the principles of passive sensing, where it does not produce radiation or ultrasound, making it safe to use. Sensing, processing, and output functions are supported by a modular architecture. To analyse signals, abdominal ECG information is preprocessed with fltering, noise removal, and normalization, and then feature extraction in time, frequency and rhythm domains. It uses a two-stage machine learning pipeline with the frst stage of the pipeline classifying the validity of the signal, and the second stage similar to a regression model clearly predicts fetal heart rate (FHR) only under the condition that a valid cardiac activity is detected. Random Forest (RF) and Bidirectional Long Short-Term Memory (BiLSTM) are tested to analyze fetal movement, with a higher accuracy of 97% of the collected data than 77% with BiLSTM. A visualization platform to enable analysis and real-time monitoring is also included in the system. In general, the presented solution proves the possibility of a low-cost, continuous, and home-based fetal surveillance with the help of wearables. The future research will be based on broadening the range of data, and better generalization of models, and attachment of clinically verifed annotations to increase reliability.