Computer aided skin cancer classification system
Skin Cancer is the most common cancer in US. Skin cancer includes melanoma and non-melanoma types of skin cancer. More than 76,000 new melanoma skin cancer cases and nearly 10000 deaths due to melanoma are estimated for 2014. In general, dermatologist have achieved correct rate of 80% for recognizing skin cancer. Early detection of skin cancer is associated with better prognosis and melanoma may be treated successfully. However, detecting the early signs of skin cancer is challenging because the cancerous structures have many features in common with normal skin tissue. To improve the recognition rate, computer aided dermoscopy (CAD) is used. There are three key techniques used for automated classification which are based on: 1) the ABCD visual diagnostic system, 2) Epiluminescence Microscopy and 3) Physics-based skin modelling. The ABCD system was originally developed for clinical use, standing for Asymmetry, Border irregularity, Color variegation and Differential structure of lesions. Several CAD skin cancer ABCD based classification systems already exists in the literature. An automatic dermoscopy image analysis system has usually five stages: (1) prepossessing step, (2) proper segmentation step, (3) feature extraction step, (4) feature selection steps, and (5) lesion classification step. The correct segmentation is the most important, since it affects the precision of the subsequent steps. The major limitations in the existing computerized algorithms are the segmentation and feature vector construction steps. The above mentioned problems can be partially solved or at least addressed considering a further important aspect in the CAD system. In this thesis, an automated ABCD skin cancer screening system has been developed, for this purposes a new segmentation method, new color space, new fractal dimension, and new hole related features are proposed. Computer simulations shows that the maximum correct classification rate up to 94.85% with sensitivity of 92.40% and specificity of 97.30% has been reached.