Optical detection and monitoring of pigmented skin lesions.

A method is presented for discriminating between malignant and benign pigmented skin lesions based on multispectral and multi-angle images. It is discussed how to retrieve maps of physiology properties and morphometric parameters from recorded images using a bio-optical model, radiative transfer calculations, and nonlinear inversion, and how to employ automated zooming to extract lesion and surrounding masks. Training and validation of a classification scheme for separation between benign and malignant tissue yielded sensitivity/specificity ranging from 97%/97% for application to a small dataset comprised of lesions not used for training and validation to 99%/93% for application to a larger dataset.