Deep Neural Networks for Heart Beat Detection

Machine learning · Biomedical signal processing · York University

Real-time heartbeat detection in noisy electrocardiogram (ECG) and ballistocardiogram (BCG) signals using deep neural networks. A densely-connected network and a two-stage CNN pipeline (presence detector + peak localizer) were trained on the MIT-BIH arrhythmia database, reaching 97.7% (dense) and 98.7% (CNN localizer) F1 accuracy on held-out test data. The work also built a data-processing pipeline for filtering and resampling ECG recordings, a model-search tool for queueing architecture experiments, and a live signal viewer for running detectors on recorded or streaming signals.

With Fasil Cheema, supervised by Yang Zhao and Dr. Peter Lian. Lassonde School of Engineering, York University.