Automatic Diagnosis of Melanoma Through the Analysis of Dermoscopic Images

Abstract

Malignant melanomas are the most dangerous type of skin cancer. It is fatal and hard to treat it, if is not treated or recognized early. Therefore, early diagnosis of skin cancer is essential to reduce mortality and morbidity of patients. The detection accuracy is also an important factor. In this paper, therefore, we perform and analytical study to investigate the importance of different handcrafted feature categories, imbalance handling methodologies and feature selection algorithms applied to melanoma diagnosis. This analysis allows us to deeply understand the importance of each feature category and to finally design a more accurate melanoma diagnosis approach. In this work, we analyze different hand-crafted based technique to investigate the effect of different features using ABCD (asymmetry, border irregularity, colour, and dermoscopic structure) rule and analyze the effect of different class imbalance handling methodologies to alleviate the effect of class imbalance problem. We applied color consistency preprocessing and scaled down all dataset images then, important features are selected from ABCD extracted features. Finally, these features are classified as benign or malignant, we founded all features group and combination of NCL and bSMOTE class imbalance handling methods produce the best result.