FitMine: automatic mining for time-evolving signals of cardiotocography monitoring

Abstract

The monitoring and assessment of the fetus condition are considered to be among the most important obstetric issues to consider during pregnancy and the prenatal period. Monitoring the fetal condition is required to detect the presence of any abnormalities in the oxygen supply to the fetus early in the antenatal or labor period. Early detection can prevent permanent brain damage and death, both of which may arise from suffocation caused by fetal disease, hypoxic-ischemic injury in the neonatal brain, or chronic fetal asphyxia. In this paper, we propose a new signal-fitting method, FitMine, that identifies the fetal condition by analyzing fetal heart rate (FHR) and uterine contraction (UC) signals that are non-invasively measured by cardiotocography (CTG). FitMine is a novel nonlinear dynamic model that reflects the relation between the FHR and UC signals; it combines the chaotic population model and unscented Kalman filter algorithm. The proposed method has several benefits. These are: (a) change-point detection: the proposed method can detect significant pattern variations such as high or low peaks changing suddenly in the FHR and UC signals; (b) parameter-free: it is performed automatically without the requirement for the user to enter input parameters; (c) scalability: FitMine is linearly scalable according to the size of the input data; and (d) applicability: the proposed model can be applied to detect abnormal signs in various domains including electroencephalogram data, epidemic data, temperature data, in addition to CTG recordings.