Novel Data Processing Approach for Deriving Blood Pressure from ECG Only

Blood pressure is one of the most valuable vital signs. Recently, the use of bio-sensors has expanded, however, the blood pressure estimation still requires additional devices. We proposed a method based on complexity analysis and machine learning techniques for blood pressure estimation using only ECG signals. Using ECG recordings from 51 different subjects by using three commercial bio-sensors and clinical equipment, we evaluated the proposed methodology by using leave-one-subject-out evaluation. The method achieves mean absolute error (MAE) of 8.2 mmHg for SBP, 8.7 mmHg for DBP and 7.9 mmHg for the MAP prediction. When models are calibrated using person-specific labelled data, the MAE decreases to 7.1 mmHg for SBP, 6.3 mmHg for DBP and 5.4 mmHg for MAP. The experimental results indicate that when a person-specific calibration data is used, the proposed method can achieve results close to a certified medical device for BP estimation.