A narrative review of methods for estimating blood pressure waveforms from photoplethysmography.

Abstract

Accurate, continuous blood pressure (BP) monitoring is important for preventing, diagnosing, and managing cardiovascular disease. Estimating the full arterial BP waveform can provide more clinical information than cuff-based measurements that report only systolic and diastolic BP. This review summarizes recent progress in predicting arterial BP waveforms from photoplethysmography (PPG) and identifies the main methodological gaps that limit clinical translation. We reviewed research published since 2021 on PPG-based BP waveform estimation. We organized the literature around core pipeline elements, including open-source datasets, preprocessing and quality control, model architectures, and validation practices. We also surveyed commercially available technologies that measure BP waveform. Data cleaning approaches can be grouped into three main categories: (1) template-based pulsation comparison, (2) feature extraction paired with classical machine learning, and (3) neural network-based quality assessment. For waveform prediction, recent work is dominated by deep learning, particularly U-Net-based architectures, autoencoder families, and attention-based models. Progress in PPG-based BP waveform prediction is rapid, but comparability and reproducibility remain major barriers. Benchmark datasets with consistent acquisition protocols, transparent preprocessing, and standardized validation criteria are needed to support fair comparison.