Automatic detection of basal cell carcinoma using vascular-extracted features from dermoscopy images

Basal cell carcinoma (BCC) is the most common type of skin cancer. Timely diagnosis of BCC is an important factor in the prognosis of the disease. A key to BCC diagnosis are vascular structures of the lesion. Detection and recognition of cutaneous vasculature provide critical information on diagnosis accuracy and assessment precision. In this paper, we present an effective method to extract vascular information towards lesion diagnosis. Given a dermoscopy image, we first segment vascular structures of the lesion by decomposing the image using independent component analysis into melanin and hemoglobin components and further applying shape filters at different scales. A vessel mask is generated as a result of global thresholding. A set of vascular features are then extracted from the final vessel image of the lesion and fed into a Random Forest classifier. The method demonstrates performance of 90.3% in terms of AUC in differentiating BCC from benign lesions.