Real-time Cuffless Blood Pressure Estimation Using Single-lead ECG and Gaussian Process Regression

Continuous blood pressure monitoring is increasingly important in wearable health applications, yet accurate cuffless estimation from single-lead electrocardiography remains challenging because blood pressure is indirectly reflected in electrocardiographic morphology, rhythm, and timing characteristics. This study proposes a feasibility oriented real-time framework for cuffless blood pressure estimation from wearable single-lead electrocardiography using Pan-Tompkins++ for robust R-peak detection and Gaussian Process Regression for systolic and diastolic blood pressure prediction. The framework integrates ECG pre-processing, beat localization, waveform segmentation, morphological and statistical feature extraction, strictly online lightweight feature-level domain adaptation, and sliding-window inference. Model training was performed using a public cuffless blood pressure dataset, while real-time evaluation was conducted using independently acquired ECG signals from 20 healthy adults under controlled seated conditions with a Shimmer wearable device. Reference blood pressure values were obtained using an Omron HEM-7120 digital sphygmomanometer validated according to the European Society of Hypertension International Protocol. Compared with the original Pan-Tompkins method, the proposed pre-processing and Pan-Tompkins++ stage improved beat detection performance, increasing sensitivity from 95.66% to 98.84% and reducing the detection error rate from 4.67% to 1.50%. In real-time window-level estimation, the proposed framework achieved mean absolute percentage error values of 3.47% for systolic blood pressure and 4.94% for diastolic blood pressure. The proposed framework achieved MAE values of 3.70 mmHg for SBP and 3.45 mmHg for DBP, RMSE values of 4.22 mmHg and 3.63 mmHg, and mean error ± standard deviation values of 2.23 ± 3.60 mmHg and 0.75 ± 3.58 mmHg, respectively. These results indicate the feasibility of wearable ECG-only blood pressure estimation in a controlled healthy-adult setting, but they should not be interpreted as clinical validation of continuous beat-to-beat blood pressure monitoring. Further validation is required in more diverse populations, wider blood pressure ranges, motion-rich conditions, and with stronger continuous reference measurements.