Computer Aided Diagnostic System for the Classification of Skin Cancer Using Dermoscopic Images

Melanoma has been recognized as the most fatal form of skin cancer to occur in humans and its incidence is increasing in many decades. The primary cause of melanoma is the over-exposure to UV rays emanating from sunlight, in those with low levels of melanin. Dermoscopy is a noninvasive procedure used for the examination of internal skin lesions. As the proposed method is based on eye-deduction, diagnosis of melanoma at early stage is quite difficult for dermatologist. This paper presents a methodological approach to extract features for benign and melanocytic skin lesions based on the ABCD rule using SVM Classifier. The image of the skin lesion is pre- processed to remove unwanted hair and noise. The lesion is segmented from the pre-processed image. Shape and colour features are extracted from the segmented lesion based on the ABCD rule. From the obtained features, it is then classified as benign and malignant using SVM Classifier. Results is demonstrated for the proposed approach has a better segmentation compared to the other techniques, obtaining values of 70% sensitivity, 75% specificity, 72.5% accuracy.