Cuffless blood pressure prediction from PPG using relevance vector machine

Blood pressure is an important parameter in monitoring and non-invasive diagnosis of cardiovascular disease. It can be estimated by analyzing photoplethysmogram (PPG) signals. In this paper, features extracted from 156 PPG signals, obtained from 26 subjects under three different conditions, relaxed, exercise and rest, are used to predict both diastolicblood pressure (DBP) and systolicblood pressure (SBP) using relevance vector machine (RVM), a supervised machine learning algorithm. In addition, support vector machine (SVM) and random forest (RF) are used to predict SBP and DBP values, and hence validate the performance of RVM by comparing their results. Cohen's kappa score is used to compare the various regression models. From the results, we can infer that RVM performs best with average kappa scores of 0.99 for both SBP and DBP, with minimum computation time.