Ranking of Different Wavelets in the Computation of Phase-Rectified Signal Averaging for Fetal Acidemia Identification

The Phase-Rectified Signal Averaging (PRSA) technique has been widely investigated for the assessment of fetal well-being during labor, through the analysis of the fetal heart rate (FHR) series. PRSA provides the average acceleration and deceleration capacities of FHR by means of synchronous average of segments associated with either an increase or decrease of the FHR. The identification of which segments to average and the computation of the capacities are currently based on the Haar wavelet at scale T and s, respectively. In this study, we proposed a generalization of the PRSA algorithm by changing the wavelet involved. We tested five different wavelets, i.e., Haar, Shannon, Morlet and two versions of Poisson's, for the identification of acidemia at birth on FHR recordings collected during labor. The ranking of the top five wavelets was created using the area-under-the-curve (AUC) of the ROC analysis. The Haar wavelet was outperformed by Shannon and Poisson wavelets, with an AUC of 0.65 and 0.60, respectively. The findings suggest that different wavelets may be more appropriate for acidemia detection.