DAJO: A Robust Machine Learning–Based Framework for Preprocessing and Denoising Fetal ECG Signals

Abstract

Accurate Fetal Heart Rate (FHR) detection and fetal electrocardiogram (fECG) analysis are vital for early identification of fetal distress. However, clinical fECG signals are often degraded by maternal ECG, baseline drift, powerline interference, and uterine contractions, reducing diagnostic reliability. To address this, the study presents a DAJO, a preprocessing framework that combines Denoising, Adaptive filtering, Joint FHR detection, and Optimized feature extraction. The workflow employs ensemble filters for noise suppression, adaptive filtering to enhance fetal-specific components, and a modified Hamilton–Tompkin’s method for robust FHR estimation. CNN-based feature extraction further ensures compact yet discriminative signal representation. Experimental results demonstrate that DAJO achieves 97% accuracy, 95% precision, 92% recall, 98% specificity, and a 95% F1 score, confirming its effectiveness. This highlights the DAJO as a robust preprocessing solution that preserves physiological integrity while improving automated FHR detection.