Classification of heart rate variability in patients with mild hypertension

Abstract

The diagnostic performance of two pattern classification methods to detect hypertension was evaluated in a population of 29 mildly hypertensive and 20 normal subjects. The heart rate variability (HRV) signal of each subject was recorded during rest and isometric handgrip exercise. Feature vectors composed of up to 6 features from both the time and frequency domain representation of the HRV signal were constructed and applied to a Bayes' likelihood classifier and a voting k-nearest neighbours classifier. Each subject was classified as hypertensive or normal, and the classification compared to the clinical diagnosis for each subject. The diagnostic performance of each classifier/feature vector combination was evaluated using the leave-one-out method. The best performance of 90% correct classifications was achieved using a nearest neighbour classifier, a Euclidean distance metric and 3 features. The Bayes' classifier achieved a best performance of 84% correct classification. The work shows promise for the detection of the autonomic disturbance which precedes and accompanies the hypertensive state.