Classification of Melanoma from Dermoscopic Images Using Machine Learning

Abstract

Malignant melanoma is one of the most vigorous and life-threatening forms of skin cancer. Mortality rates of melanoma are quite high because it can easily proliferate to other parts of body. Despite such dreadful facts, melanoma can be cured if diagnosed at initial stage. So, it is significant to differentiate melanoma and benign lesions. For early diagnosis, CAD systems can prove worthful as they do not demand invasive procedures. In this work, methodology is proposed for classification of melanoma and non-melanoma. Initially, pre-processing steps are executed to cleanse and enhance dermoscopic images by removing undesirable objects. Then, skin lesion which is of major region of interest is obtained using active contour-based segmentation. Furthermore, colour, texture and shape features are extracted and given to SVM classifier for effective and efficient classification of melanoma and benign lesions. Experimental results are acquired using PH2 data set, and it has been seen that our proposed method yields accuracy, sensitivity and specificity of 94.5%, 82.5% and 97.5%, respectively.