Robust algorithm to locate heart beats from multiple physiological waveforms

Alarm fatigue is a major issue in patient monitoring that could be reduced by merging physiological information from multiple sensors, minimizing the impact of a single sensor failing. We developed a heart beat detection algorithm that utilizes multi-modal physiological waveforms (e.g. ECG, blood pressure, stroke volume, photoplethysmogram and electroencephalogram). The 100 record training set from the Physionet challenge 2014 was used for development. The algorithm was evaluated at three testing phases during the 2014 challenge consisting of 100 (phase I), 200 (phase II) and 300 (phase III) hidden records, respectively. A true positive was declared if a beat was detected within 150 ms of a reference annotation. The algorithm had a sensitivity of >99.9%, Positive Predictive Value of 99.7%, and an overall score (average of sensitivity and Positive Predictive Value) of 99.8% when applied to the training set. The best overall performance on the test sets were: 88.9%, 76.3% and 84.4% for phases I, II and III, respectively. We developed a robust heart beat detector that fuses annotations from multiple individual detectors. The algorithm improves the training results compared to ECG detections alone, and performs well on the test sets. Data fusion approaches like this one can improve patient monitoring and reduce false alarms.