OSGAN: Omni-scale and Global-aware ECG arrhythmia diagnostic network

Chao Chen, Bo Jin, Chao Che, Ruixin Li

Abstract: Highlights•Overlapping-group for hierarchical feature extraction of 12-lead ECGs.•Cross-Fusion enhances the global perception capability of CNNs.•An attention mechanism using a long-term dependency mask for long dependency modeling of ECGs.•OS-CNN block with SE module which enables feature weighting for different receptive fields.