Robust heart rate estimation from multimodal physiological signals using beat signal quality index based majority voting fusion method

Abstract

In this paper, we present a new beat signal quality index (SQI) based majority voting fusion algorithm for robust heart rate (HR) estimation from multimodal physiological signals, namely, cardiovascular and non-cardiovascular signals. A novel statistical and probabilistic based beat SQI assessment method has been developed for voting fusion. Modified slope sum function and Teager-Kaiser energy operator method has been used for beat detection in electrocardiogram (ECG) and non-cardiovascular signals. The performance of majority voting fusion method in beat detection has been evaluated on PhysioNet/CinC Challenge-2014 public training dataset and has achieved overall score of 94.93%. The performance of the algorithm has been tested on PhysioNet/CinC Challenge-2014 hidden test set and MIT-BIH Polysomnographic dataset and it has achieved scores of 90.89% and 99.77% respectively. The proposed method has improved average�rMSE�of HR estimate from 15.54�bpm to 0.24�bpm for noisy ECG signals and from 11.68�bpm to 0.84 bpm for noisy ECG and noisy ABP signals of PhysioNet/CinC Challenge-2014 public training database. The majority voting fusion method has yielded HR estimate with average rMSE of 1.80�bpm, when both ECG (avg. rMSE of 4.58�bpm) and ABP (avg. rMSE of 3.96�bpm) signals of MIT-BIH Polysomnographic dataset are noisy. The use of multimodal signals in fusion has increased the accuracy of HR estimates in noisy ECG and ABP signals. The majority voting fusion algorithm based on beat SQI has enabled effective and reliable use of non-cardiovascular signals in robust HR estimation from multimodal physiological signals, even when both ECG and ABP signals are noisy.