Blood Pressure Estimation from PPG: A Comparative Study of Direct and ECG-Mediated Deep Learning Pipelines

This preprint compares direct PPG-to-blood-pressure prediction with pipelines that first reconstruct ECG from PPG and then estimate blood pressure. In a physiological correlation analysis of the MIMIC-III waveform database, PPG showed stronger coupling with arterial blood pressure than ECG. Across 1.74 million segments from 3,127 patients, the direct pipeline achieved British Hypertension Society Grade A performance, with SBP and DBP mean absolute errors of 4.82 and 4.31 mmHg, respectively. ECG-mediated approaches reached only Grade B accuracy, suggesting that ECG may add complexity without improving estimation.