Cuffless Photoplethysmogram Based Beat-Beat AI Powered Continuous Blood Pressure Estimation Towards Precision Medicine

Continuous blood pressure (BP) monitoring is crucial for cardiovascular management, but traditional cuff-based systems are limited in practicality and real-time usability. This study presents an AI-driven approach utilizing photoplethysmography (PPG) signals from 219 subjects in the PPG-BP Dataset. From each subject's PPG waveform, features were extracted across five signal domains: time, frequency, entropy, recurrence, and statistical. Using SelectKBest for domain-specific feature selection, the top 5 features from each domain were identified and combined to form a unified feature set. Multiple regression models - including Huber, TheilSen, SVR, RANSAC, and Linear Regression - were trained to predict systolic (SBP) and diastolic (DBP) blood pressure. The best-performing models using the combined feature set were the Huber Regressor for$\mathbf{S B P}\left(\mathbf{M A E}=\mathbf{0. 2 0 7 9}, \mathbf{R}^{\mathbf{2}}=\mathbf{0. 9 9 5 9}\right)$and the TheilSen Regressor for DBP (MAE$\left.=\mathbf{0. 1 7 3 0}, \mathbf{R}^{\mathbf{2}}=\mathbf{0. 9 9 6 4}\right)$, significantly outperforming other methods. This domain-wise hybrid modeling approach demonstrates strong potential for accurate, interpretable, and non-invasive BP estimation, enabling real-time monitoring through wearable AI-based systems towards precision medicine.