Computer-Aided Diagnosis Method for Paroxysmal Atrial Fibrillation based on Cardiac Dynamics Underlying Surface ECG Signals

Paroxysmal atrial fibrillation (PAF) is a common arrhythmia characterized by irregular and episodic atrial contractions, whose unpredictability and intermittent nature pose significant health risks to individuals. Early detection of PAF is crucial for timely intervention and effective management. However, the scarcity of medical data for this condition presents a significant challenge, making it difficult to establish reliable classification models. This paper proposes a new classification method based on 12-lead non-epiodic ECG signals by combining cardiac dynamics features with one-class classification algorithm for PAF detection. Three-dimensional dominated components are extracted to reduce the complexity of the ECG data while retaining essential features. And deterministic learning mechanism is adopted for cardiac dynamics features extraction to reflect complex system patterns of nonlinear ECG signals. By applying spatial information measures to dynamical features, this serves as an input to one-class classifiers for the final PAF detection. The method is able to detect PAF even before pathological changes in the surface twelve-lead ECG signal.