Development of peak detection algorithm using window variance for physiological variability monitoring

Physiological variability has gained importance in the assessment of autonomic function as well as monitoring of severity and prognosis of various diseases in last 5 decades. Electrocardigram or peripheral arterial pulse is generally used for deriving variability spectrum. Presently manual processing and analysis of these signals are done in the absence of a rugged peak detection method yielding no false positive with few false negatives. Continuous patient monitoring demands automatic peak detection without any manual intervention. False positive peak detection can result in erroneous variability spectrum, though false negatives can be interpolated within limits (1-5%). With above in view, an algorithm named window variance is proposed which uses a moving window of the input signal (0.3 seconds) with a shift of 0.025 seconds to locate peaks and use variance to eliminate false positives. This algorithm has been tested in 5 minutes of continuous data (with and without superimposed random noise) from 5 volunteers, comprising 1907 true peaks yielding no false positives and very few false negatives (0.1573%).