Skin characterization by ultrasound techniques

Ultrasounds have an increasingly important role in the diagnosis and monitoring of skin, since the used techniques are non-invasive and non-ionizing, and also provide internal images of the skin morphology over the depth. To evaluate the power of ultrasounds to help in the lesions diagnosis, first some phantoms were developed trying mimicking the human skin tissue. Thus, several gelatin-based phantoms have been constructed with scattering agents as flour, starch and glass microspheres. Then, A-Scan lines were extracted using the pulse echo approach, followed by the ultrasound velocity. The obtained results for the considered acoustical parameter present good concordance with the ones found in literature. Methodologies were developed to analyze the using phantoms and human images with and without lesions. Two approaches were considered: (1) the identification of the superficial skin boundary working as a skin lesion discriminator, (2) and binary classification, characterizing the presence/absence of lesions, using several regions of interest by the Support Vector Machine and the Naive Bayes classifiers.