Diagnosing the major contributing factors in the classification of the fetal health status using cardiotocography measurements: An AutoML and XAI approach

The universal criticality of mother’s and child’s health can scarcely be overstated. Deterioration in an expectant mother’s health may lead to several complications in the antepartum and intrapartum period that which may be fatal. Hence, simultaneous tracking of prenatal parameters such as uterine contraction and Fetal Heart Rate (FHR) through Cardiotocography (CTG) is of critical importance. This preponderantly remains a manual procedure in developing nations as the inclusion of machine learning (ML) technology is still not widespread. Numerous studies have pointed to the fact that electronic monitoring of FHR has not had any noteworthy benefit in lowering the incidence of perinatal mortality and morbidity. Further, FHR monitoring is beset with other problems such as high disparities in intra-and inter-observations and enhanced rate of caesarean vis-à-vis normal delivery. These drawbacks of FHR notwithstanding, the fact remains that FHR is regarded as a vital obstetric procedure and CTG as equipment is deployed the most in perinatal diagnostics. With advancements in medical technology, techniques viz. computerized analysis of FHR and ECG are now being utilized as an accompaniment to CTG. These sophisticated computerized methods enable an analysis of variations in heart rate and assessing the patterns of short term heart rate. Further studies about multiple parameters causing FHR variability is needed to better our understanding of the primacy and salience of various parameters of raw fetal heartbeat data. Our experimental results using AutoML approach gave an accuracy rate of 0.9561, Recall 0.9056, Kappa 0.8792, Precision 0.9552, AUC 0.9864, F1 0.9550, MCC 0.8805 with model handling time of 2 minutes.