Sinus rhythm heart rate estimation in high noise environments by application of a priori RR interval statistics.

Most existing heart beat detection algorithms serially process peaks, which can be either noise or true beats. Serial processing can result in inaccurate detections in the context of high noise. The proposed method relies on the relative regularity of sinus rhythm RR interval changes to select the best sequences of peaks in a 5-10 s long segment of cardiac data. The best sequences with a current data segment are subjected to a trending analysis, to determine whether their associated RR intervals fit within a pattern of prior best segments. The RR regularity scores and the results of the trending analysis are combined into a single sequence score and the final sequence for a segment is chosen from the best sequences based on this overall score. The current heart rate estimate is updated with the final sequence's RR interval by an adaptive filter that weights the overall score. Twenty-four hour RR interval records for 54 normal individuals were parsed into 10-s segments and corrupted with spurious 'noise' peaks, which resulted in a revised RR interval series that included a number of false RR intervals. The algorithm was run on these corrupted RR interval series. The percentages of mean heart rate values within 5 beats min(-1) of the true value were 95%, 88% and 77% for 10, 20 and 30 added noise spikes, respectively. The percentages of mean heart rate values within 10 beats min(-1) of the true value were 98%, 96% and 91% for 10, 20 and 30 added noise spikes, respectively. Accuracy was higher for data segments characterized by relatively low RR interval variability. The proposed algorithm shows promise for estimating average heart rate for sinus rhythm in high noise environments.