Color analysis in boundary areas of skin tumor images for melanoma diagnosis

This thesis focuses on the development and implementation of color analysis of melanoma images to make a diagnosis. Melanoma is a form of skin cancer, which can be deadly if not treated in the early stages. Current melanoma detection techniques rely on guidelines, like asymmetry, border irregularity, color and diameter of the tumor, called the ABCD criteria [1]. Among them, color analysis becomes an important index in the diagnosis of malignant melanoma. It is also hypothesized that the clustering of melanoma-colored pixels occurs more intensely in melanomas than in benign lesions. It is believed that melanoma colors and color clustering may have some particular characteristics in the boundary region of malignant melanoma. The main objectives of this research are to determine if there are some special features in the boundary region of the melanoma, and to try to improve the diagnostic accuracy based on its features. By processing 50 clinical melanoma images and 50 clinical benign images, the research finally indicates that the diagnostic result is more encouraging if we analyze the image based on melanoma colors or clustering in the offset boundary that only encompasses a certain percentage of the total tumor size --Abstract, page iii