Blood pressure (BP) is among the most important vital signals. Estimation of absolute BP solely using photoplethysmography (PPG) has gained immense attention over the last years. Available works differ in terms of used features as well as classifiers and bear large differences in their results. This work aims to provide a machine learning method for absolute BP estimation, its interpretation using computational methods and its critical appraisal in face of the current literature. We used data from three different sources including 273 subjects and 259,986 single beats. We extracted multiple features from PPG signals and its derivatives. BP was estimated by xgboost regression. For interpretation we used Shapley additive values (SHAP). Absolute systolic BP estimation using a strict separation of subjects yielded a mean absolute error of $9.456\mathrm{mmHg}$ and correlation of $0.730$. The results markedly improve if data separation is changed (MAE: $6.366\mathrm{mmHg}$, r: $0.874$). Interpretation by means of SHAP revealed four features from PPG, its derivation and its decomposition to be most relevant. The presented approach depicts a general way to interpret multivariate prediction algorithms and reveals certain features to be valuable for absolute BP estimation. Our work underlines the considerable impact of data selection and of training/testing separation, which must be considered in detail when algorithms are to be compared. In order to make our work traceable, we have made all methods available to the public.

Imaging photoplethysmography allows to capture spatio-temporal patterns related to the perfusion. One such approach is based on the analysis of the time delay between pulse waves at different locations by so-called phase maps. There are different ways to establish such maps. However, neither a comparison between existing methods has been published nor has the impact of different stimuli been sufficiently examined until today. In this work, we compare three previously published approaches for the generation of phase maps and investigate the impact of two physiological stimuli on such maps. Our results show pairwise correlation coefficients between the different approaches of phase map generation from r = 0.65 to r = 0.82, indicating substantial differences between maps. The different maps reflect our physiological expectation in varying degrees. Particularly for a weaker (distant) stimulation refinements are needed to reveal characteristic changes.

Assessment of skin perfusion can reveal signs of deterioration and help to prevent critical states. Commonly applied clinical tests to capture skin perfusion are subjective and often hard to quantify. Laser speckle contrast analysis (LASCA) is a technique that can capture skin perfusion at high spatial and temporal resolution. LASCA requires, however, a complex setting and is not suited for monitoring under clinical conditions. Imaging photoplethysmography (iPPG) might be an easy to use alternative. However, direct comparisons of LASCA and iPPG, and thus proves of iPPG’s capabilities to capture skin microperfusion, are rare. In this work, we compare the longitudinal development of the amplitude of green channel and near-infrared iPPG and LASCA after application of a hyperemic test. Our results show statistically significant increases in amplitude over time in all modalities. The maximum increase in median normalized amplitude is 1.281 for the green channel, 0.594 for near-infrared and 1.111 for LASCA. Median Spearman’s rank correlation coefficient of the amplitude for green channel and LASCA is r= 0.89 and r= 0.71 for near-infrared and LASCA.