Comparative evaluation of on-line missing data regression techniques in intrapartum FHR measurements

Cardiotocography is a measurement technique used for fetal health status assessment in antepartum and intrapartum monitoring. A typical cardiotocogram recording consists of two simultaneously acquired signals, namely the fetal hearth rate (FHR) and the uterine activity (UA). Unfortunately, the FHR recordings suffer from frequent invalid or missing samples, due to artifacts or sensors misfunctions. In literature, this problem is typically solved by simplistic linear interpolations or sophisticated algorithms, whose computational complexity prevents from an on-line implementation. In this paper, we propose five regression techniques which rely exclusively on a reduced set of past samples, i.e. compliant with on-line implementation. We characterize their performances in terms of regression error, as well as specific clinical indices which account for the capability not to distort the original information content of the acquired signal.