Classification of Melanoma using different segmentation techniques

Abstract

Malignant melanoma is the most dangerous type of skin cancer which has capability to spread to other parts of body and proves deadly for a person. Mortality rates of melanoma are quite high. Also, number of melanoma patients is inclining at a faster pace. In such scenarios, Computer aided systems can act as savior because early detection of melanoma increases chances of survival. These CAD systems are based on image processing techniques. Among these techniques, Segmentation is the most challenging task because melanoma moles appear in random patterns having ragged and irregular shapes. In this work, classification of melanoma and non-melanoma is performed using four different segmentation techniques namely otsu, k-means clustering, maximum entropy and active contour individually. Comparison of results obtained using all these techniques has been carried out and it is seen that among these techniques active contour-based method proves eminent in efficient classification of melanoma and non-melanoma by showing accuracy of 94.5%.