Automatic Deceleration Detection System from Fetal Heart Rate obtained by CTG

The purpose of this study is to detect decelerations automatically from the fetal heart rates obtained by cardiotocography (CTG). CTG records fetal heart rates in order to identify high risk fetuses, allowing appropriate obstetrical interventions. Deceleration, one of the features of fetal heart rates observed by CTG, is a temporary decrease in heart rates, which can be an indicator for adequate interventions. Deceleration detection based on visual observation of CTG has low inter-rater reliability, especially in illegible CTG traces. This problem may lead to undesirable obstetrical interventions. Therefore, there is needs of methods to detect decelerations, providing reproducible and objective interpretation that are difficult by the human eye. In this paper, we propose an algorithm to automatically detect decelerations from illegible CTG traces using Singular spectral analysis (SSA) and the Hilbert transform. CTG traces include a total of 564 decelerations. As a result, the proposed method achieved 84.0%, 89.4% and 86.7% of sensitivity, positive predictive value and F1 score, respectively. These results indicate superior performance of the proposed method compared to conventional algorithm. This method is used only off-line manner, and is expected to further improve detection accuracy and be applied to real-time analysis.