This paper proposes a real-time approach for validating heartbeat segmentation using convolutional neural networks (CNNs). The method aims to reduce false positive alarms by classifying heartbeat patterns as either 'heartbeat' or 'not a heartbeat'. A seven-layer CNN is used, and the approach is evaluated on the MIT-BIH and CYBHI databases. Compared to the Pan-Tompkins algorithm, the CNN approach improves positive prediction while slightly decreasing sensitivity, demonstrating feasibility for real-time applications and potential for embedding in dedicated hardware.

Pedro Silva, Eduardo Luz, Guilherme Silva, Gladston Moreira, Elizabeth Wanner, Flavio Vidal, David Menotti