Automated fetal heart rate analysis for baseline determination and acceleration/deceleration detection: A comparison of 11 methods versus expert consensus

Abstract

The fetal heart rate (FHR) serves as a guide to fetal well-being during the first stage of delivery. The visual morphological analysis of the FHR during labor is subject to inter- and intra-observer variability – particularly when the FHR is abnormal. It has been suggested that automatic analysis of the FHR can reduce this variability. To compare 11 morphological FHR analyses (baseline computation, and detection of FHR decelerations and accelerations) produced by automatic analysis methods (AAMs) with an expert consensus. Eleven AAMs were reprogrammed (using the description published in the literature) and applied to 90 FHR recordings collected during the early phase of labor. Furthermore, the recordings were divided into three tertiles, according to the difficulty of analysis. The results of the morphological FHR analyses produced by the AAMs were compared with a consensus morphological analysis performed by four experts. In addition to standard discriminant criteria, a new morphological analysis discriminant index (MADI) was introduced; it provides an overall evaluation that collates all the individual criteria. The AAM developed by Lu and Wei's gave better results than the other AAMs for baseline computation. Regarding this method's detection of FHR decelerations and accelerations, the F-measure [95% confidence interval] was respectively 0.73 [0.67; 0.76] and 0.70 [0.64; 0.76]. The MADI indicated that Lu and Wei's AAM agreed best with the expert consensus (discordance: 7.3% [6.10; 8.60]). Our study demonstrated the superiority of Lu and Wei's method for baseline computation and deceleration/acceleration detection, although there was still a significant degree of discordance versus expert consensus. The MADI appears to be a good overall index for evaluating AAMs with regard to the quality of baseline computation and acceleration/deceleration detection. The application of precise criteria and the methodology and software tools developed here should facilitate the evaluation of new AAMs and their comparison with other methods.