Fig. 1: Study workflow.

a, RSOM scanning setup. The controller regulates the motion of the stages. b, RSOM principle of operation. The handheld probe with mounted illumination is fixed on the skin and the motorized stages move the ultrasound detector along the XY plane over the scanned skin region. For each light pulse, a number of ultrasound waves are produced. ED, epidermal layer; DE, dermal layer. c, The ultrasound detector records an ultrasound 1D signal at each scanning point in a single-pulse-per-signal manner. d, Recorded ultrasound signals are reconstructed into a volumetric 3D RSOM image. The different colours represent different frequencies (green, high frequencies; red, low frequencies). The white dotted lines represent the limits of the epidermal/SVP layer. e, Finally, a tomographic 2D MIP (along the Y axis) RSOM image is produced. Scale bars, 500 μm. f, After segmenting the microvasculature in the RSOM images, we extracted 64 geometry-related skin features. Next, a cxAI approach identified the 32 most diabetes-associated features which were further related to diabetes stage/progression.