Event-based sampled ECG morphology reconstruction through self-similarity.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Event-based analog-to-digital converters enable sparse bio-signal acquisition but can lose critical bio-markers.
• Standard interpolation techniques struggle to recover features lost during aggressive event selection in ECG signals.

Purpose of the Study:

• To leverage the self-similarity of ECG signals for recovering missing features in event-based sampled data.
• To develop a novel dynamic time warping algorithm for inferring heartbeat morphology from sparse samples.
• To improve the accuracy of bio-marker detection in sub-Nyquist sampled ECG.

Main Methods:

• Acquired uniformly sampled heartbeats and used graph-based clustering to define patient-representative templates.
• Matched each event-based sampled heartbeat to its morphologically nearest template.
• Reconstructed heartbeats using piece-wise linear deformations guided by a novel dynamic time warping algorithm.

Main Results:

• Synthetic tests on a large dataset of normal heartbeats demonstrated significant performance improvements over standard resampling.
• Achieved up to a 10-fold improvement in P-wave detection compared to classic linear resampling.
• Showcased up to a 3-fold improvement in T-wave detection and a 30% enhancement in dynamic time warping morphological distance.

Conclusions:

• Developed an event-based processing pipeline utilizing signal self-similarity for ECG signal reconstruction.
• Demonstrated clear advantages of the proposed method over classical resampling techniques through synthetic testing.
• The approach shows promise for more robust analysis of sparsely sampled bio-signals.