Evaluation of floatingline and foetal heart rate variability

Abstract Foetal heart rate (FHR) variability is known to be a very important feature in the diagnosis of foetal well-being. Despite its clinical importance and the widespread use of foetal monitoring, a standard definition of FHR variability (FHRV) and an agreement concerning the methodologies to be employed in its evaluation are still lacking. Often, FHRV is computed in tracts of FHR signals in which both accelerations and decelerations are absent, thus making it very difficult to assess it for signals with several and closely spaced events of this kind. In this work, we propose an automated method for estimating the FHRV signal, defining it as the difference between the FHR signal and the floatingline, where the latter is the imaginary line that follows accelerations and decelerations, taking into account the frequency characteristics of these events. We tested the software developed for this purpose on both simulated and real FHR signals (sets of 50 signals). In the case of simulated signals, the average value of the mean square error vector between the simulated floatingline and that estimated was only 0.04 bpm 2 . In the case of real signals, however, in absence of a reference gold standard, the estimated floatinglines were visually assessed by a team of five expert obstetricians who judged them matching to the definition in 96% of cases. As regards the evaluation of FHRV, using the simulated FHR signals, we compared the estimated values with the reference values of short term variability (STV) and sympathovagal balance (SVB), two very significant parameters employed in computerised foetal monitoring, and obtained an error lower than 1.5% for the STV index, and an underestimation of the SVB index with an error of about 4.5%. Finally, we compared the proposed method for the estimation of the floatingline with more traditional filters (moving average and FIR with Hamming window) which showed, on average, a worse performance (quantified by mean square errors up to five times higher).