Real-time, data-driven system to learn parameters for multisite pacemaker beat detection

While beat detection in a Cardiac Electrical Signal (CES) is a well-studied problem, we propose a novel algorithm for implementation in pacemakers that learns the heights and widths of atrial and ventricular peaks from processing a few seconds of cardiac data sampled at 1 kHz. This purely data-driven solution to learn the parameters of atrial and ventricular peaks will allow pacemakers to set their own detection parameters and adaptively tune the parameters over time for that specific patient. We have validated our algorithm on one minute of cardiac electrical signal data from 51 separate channels, coming from 17 electrodes connected directly to cardiac tissue on each of three separate animals. Additionally, we have implemented the algorithm on a Field Programmable Gate Array and tested it on an ex-vivo rat heart to illustrate that our algorithm can run in real-time on commodity hardware.