Accounting for skin pigmentation in clinical photoacoustic imaging using data-driven methods

Abstract:: Multispectral photoacoustic imaging (PAI) enables non-invasive estimation of physiological biomarkers such as blood oxygen saturation (sO₂), but its accuracy can be affected by epidermal melanin. Skin pigmentation alters the optical fluence reaching deeper tissue and generates strong superficial photoacoustic sources, producing spectral coloring and acoustic clutter that distort reconstructed signals. In this work, we investigate the physical basis of these skin tone-related effects, controlled forward modelling, and clinical observations. The forward model shows that increasing epidermal melanin biases vessel sO₂ estimation despite unchanged underlying physiology, while healthy volunteer data from 42 healthy volunteers across all Fitzpatrick skin types demonstrate similar trends in clinical study. These findings show that skin tone effects in multispectral PAI arise from coupled optical and acoustic mechanisms and highlight the need for data-driven correction methods that account for non-local signal formation, skin-tone diversity, and physical interpretability.

Themes