Identification of a signal for an optimal heart beat detection in multimodal physiological datasets

This work describes an algorithm for the robust detection of heart beats in multimodal physiologic data, developed for the PhysioNet/Computing in Cardiology Challenge 2014. Depending on which physiological signals were available in the provided datasets, the proposed algorithm uses a combination of the ECG, the continuous blood pressure (BP) or the stroke volume (SV) signals. Due to the temporal dynamics of the signal distortions, each record was divided into several subsegments of the same length. Different peak detection algorithms were applied to the different signals of each subsegment. Each signal was rated with a quality index. It was used to identify one signal to be used for the heart beat detection. The quality index was estimated from the signal statistics and the number of peaks and their location within each subsegment. Once each signal of a subsegment was rated with the quality index, the best rated signal was considered for the final peak detection. This identification procedure was then repeated for every new subsegment. In the challenge, the proposed method achieved an overall score of 90.04% in phase I, 83.79% in phase II and 84.31% in phase III.