Skin capillary extraction technique based on independent component analysis and Frangi filter using videomicroscopy.

Abstract

Videomicroscopy can be used to observe skin capillaries easily and non-invasively. In this study, we develop an algorithm that can handle skin regions by combining color component separation methods as pre-processing based on blood vessel extraction filtering. Images of skin on the inner upper arm were acquired using videomicroscopy. An algorithm using independent component analysis (ICA) and the Frangi filter was constructed, and capillary regions were extracted. The capillary blood vessel extraction images were compared with ground truth to verify accuracy. An evaluation of the physiological responses of skin exposed to carbon dioxide (CO2 ) water, local heating, and methyl nicotinate was performed to test blood vessels with different mechanisms of action and layer depth. Based on a comparison with ground truth images, a Dice coefficient of 0.82 was calculated. In examining physiological responses to stimuli on the skin, it was found that exposure to CO2 for 2 minutes significantly increased the capillary area compared with the steady state. An algorithm to extract capillaries from skin images using ICA and the Frangi filter method was proposed. Results suggest that this algorithm can quantitatively analyze physiological changes in capillaries on the skin surface.