Learning-Based Model for Central Blood Pressure Estimation using Feature Extracted from ECG and PPG signals.

Pre-detection of hypertension mostly considers the measurement of Brachial Artery Blood Pressure (BABP). Although being a standard vital, it is still considered a poor alternative for Central Blood Pressure (CBP). However, CBP is measured invasively during the process of cardiac catheterization (Cath). Though cuff-less techniques to estimate BABP are widely employed, CBP estimation has not been explored yet. Moreover, to best of our knowledge intermittent CBP estimation has not been proposed earlier. Therefore, we present a cuff-less and beat-by-beat CBP estimation technique using linear regression analysis on features extracted from continuous Electrocardiogram (ECG) and Photoplethysmograph (PPG) signals. Unlike for BABP estimation, 30 supplementary features to conventional pulse transit time such as ST-interval, Psystolic peak interval, etc., were extracted to enhance CBP accuracy. This extraction was done using Haar wavelet along with modulus maxima. Feature selection has been done using the wrapper technique and reduced using principal component analysis. Segregation of each beat was achieved with the help of constraints developed based on iteration and backtracing. This model estimates Systolic CBP with a validation error of 0.109±2.37 mmHg and Diastolic CBP with an error of 0.031±2.102 mmHg for 33 Cath lab patients.