Abstract
Blood pressure measurement and continuous control are essential for heart and blood pressure patients. Therefore, continuous blood pressure measurement from these patients is required. In this paper, a novel hybrid prediction method combining Gaussian process regression (GPR) and feature extraction stage has been proposed and then applied to the estimation of blood pressure from cuff Oscillometric waveforms (or signals). First of all, 27 features including time, chaotic, and frequency domain have been extracted from these waveforms to detect the systolic blood pressure (SBP) and diastolic blood pressure (DBP) automatically. As the second stage in the proposed method, three different GPR methods comprising Exponential GPR, Matern 5/2 GPR, and Rational Quadratic GPR, have been used to estimate the SBP and DBP values based the extracted 27 features. As the performance measures, we have used seven different metrics including MAE (mean absolute error), MSE (mean square error), RMSE (root mean square error), R2, IA (index of agreement), and MAPE (mean absolute percentage error) for evaluation of the proposed methods concerning estimation performance of SBP and DBP values from cuff Oscillometric waveforms. The obtained MAPE values for Exponential GPR, Matern 5/2 GPR, and Rational Quadratic GPR in the estimation of SBP values from cuff Oscillometric signals are 0.1136, 0.2286, and 0.1745, respectively. The obtained MAPE values for Exponential GPR, Matern 5/2 GPR and Rational Quadratic GPR in the estimation of DBP values from cuff Oscillometric signals are 0.2878, 0.4220, and 0.4150, respectively. The experimental results have demonstrated that the best-proposed hybrid model is the combination of the Exponential GPR and the feature extraction stage for the estimation of both SBP and DBP values. The proposed method could be safely used in the medical blood pressure measurement systems in the hospital and clinics.