Fetal Movement Detection by Wearable Accelerometer Duo Based on Machine Learning

Objective: Fetal movement counting is one of the most important indices reflecting the health of the fetus. In the hospital, ultrasound method serves as gold standard but could only be used within a short time because of the potential of tissue damage. Out of the hospital, maternal perception is recommended by obstetrician-gynecologist while its accuracy varies between pregnant women and typically is low. This study aims to develop a wearable device for out-of-hospital monitoring of fetal movement with acceptable accuracy.
Methods:
In this work, a wearable device with two accelerometers was developed for accurate and continuous fetal movement monitoring. Nine different machine learning algorithms with optimized hyperparameters and features were compared and analyzed, and the ensemble method was found to have the best performance compared to ultrasound as gold standard on 20 pregnant women. Feature importance factor and optimized number of accelerometers were also studied.
Conclusions:
The high-performance metrics indicate that our system could potentially replace maternal perception for accurate out-of-hospital fetal movement detection. Furthermore, because it’s intrinsically passive and wearable, it could broaden the application horizon of fetal movement monitoring, and be applied to discover more rich biological profile of fetus during perinatal period.