A Hybrid deep learning framework for FHR abnormality detection

Abstract

Observing the fetal heart rate (FHR) during pregnancy and labor is a crucial tool for assessing the health of the baby in womb. High individual variability, unequal class sizes, and minor timing problems can make it challenging to understand FHR patterns. The core objective of this work is to propose a hybrid artificial intelligence framework that combines LSTM-based temporal embeddings and XGBoost classification. For reducing class imbalance, synthetic FHR data is generated using a Conditional Tabular GAN (CTGAN). The significant features including mean FHR, variability metrics, accelerations, and decelerations, are identified by the feature selection. Moreover, feature selection also used automatically learned embeddings from raw FHR sequences. From the experimental results, it has been proved that the proposed hybrid framework yields the highest accuracy as 99.51%. Precision, recall, F1 score and accuracy are the obtained results were evaluated. The proposed framework is more useful in identifying the irregular patterns in FHR. In future, the real time monitoring system will be integrated with this framework would assist health care professionals in continuous fetal monitoring and early detection of abnormalities in FHR and other organs also.

Ähnliche Arbeiten