Fetal Movements Identification Based on Non-negative Matrix Factorization and Spectral Clustering

Fetal movement patterns are a measurement of fetal well-being and it is important to ascertain fetal movements to avoid fetal morbidity and death. Fetal movement counting by mother is an unstructured method, used to quantify her baby's movements. This highly depends on her personal experience and level of attention which may not be accurate. This research is focused on developing an algorithm to auto count fetal movements which can be combined with a domestic and wearable device. Extracting the desired accelerometric fetal signals from the abdomen is arduous due to the presence of additional interference, such as mother's respiration, movements, cough and laugh signals which should be diminished. These obstacles were conquered using Dynamic time warping, Spectral Clustering and Non-Negative Matrix Factorization (NMF) based techniques. Fetal movement separation strategy in time domain was accomplished using dynamic time warping and spectral clustering. Further, this analysis was extended to frequency domain, utilizing the NMF based algorithm. In this paper, we explore those both algorithms and ensure a feasible method to separate fetal movement signal by comparing the results.