Non-Invasive Fetal Electrocardiogram Arrhythmia Classification Using RR-Interval-Based Feature Extraction and Wrapper-Based Feature Selection
Аннотация
Fetal arrhythmia is an abnormal heart rhythms that lead to adverse outcomes, including sudden infant death, making early detection crucial. Non-invasive fetal electrocardiography (NI-FECG) is a safe and promising technique for assessing fetal cardiac health by analyzing abdominal ECG signals obtained through electrodes placed on the maternal abdomen. Although substantial progress has been achieved in adult ECG analysis, NI-FECG remains a significant challenge due to signal interference and low signal-to-noise ratios. Existing studies using computer-aided algorithms have achieved high accuracy in fetal arrhythmia detection, yet they often overlook RR-interval analysis, a critical marker of heart rate variability and fetal well-being. This study proposes a machine learning-based approach for fetal arrhythmia classification using time-domain statistical features extracted from RR-intervals. Teaching-Learning-Based Optimization (TLO), a wrapper-based feature selection method, was employed to enhance classification performance and obtain the optimal feature. Using K-Nearest Neighbor (KNN) with TLO, the optimized model achieved outstanding accuracy, sensitivity, specificity, and precision in test sets (unseen). Its results highlight the effectiveness of combining RR-interval-based feature engineering with intelligent optimization for accurate fetal arrhythmia detection.