Fetal Movement Identification from Multi-Accelerometer Measurements using Recurrent Neural Networks

This work presents two approaches, the many-to-many method and the many-to-one method, to identify fetal movements, especially fetal kicks, using recurrent neural networks from multi-accelerometer readings acquired from four accelerometers mounted on the abdomen of a pregnant woman. Additionally, we discuss some issues associated with the imperfect ground-truth labelling, acquired through maternal perception, that deteriorate and misrepresent the network performance.