Generalizable deep learning for photoplethysmography-based blood pressure estimation — A benchmarking study

Summary

This paper by Moulaeifard, Charlton, and Strodthoff at Oldenburg and Cambridge provides the first rigorous out-of-distribution (OOD) benchmarking study for PPG-based blood pressure estimation. The authors train five representative deep learning models on the large-scale PulseDB dataset (MIMIC-III + VitalDB, 5.24M segments) and evaluate their in-distribution (ID) and OOD performance on multiple external datasets. A key finding is that ID performance — the only metric reported by most published PPG-BP papers — is systematically over-optimistic and does not predict OOD performance.