Melanoma Segmentation Based on Multi-stage Approach Using Fuzzy and Graph-Cuts Methods

Abstract

Globally, skin cancer is one of the major health problems. While an early diagnosis with the proper management of the disease can successful help in the treatment of the disease, the assessment of the disease by a medical practitioner is time-consuming, subjective, and prone to bias due to variation in the training and experience of dermatologists. Although different automated methods of the disease’s diagnosis have been proposed, various problems like image noise due to varying illumination, uneven low contrasts and ambiguities in the non-diseased skin and tumours in the different regions of the clinical image alongside with the edges and boundaries have been highlighted to require accurate discrimination during the use of the automated methods. This is due to the fact that they can lead to inaccurate extraction of the melanoma skin cancer in the medical images as the problems plague the performance of an automated approach of detecting the disease. This study implements a multi-stage image segmentation approach that utilises a fuzzy transformation at the image enhancement stage with graph-cuts technique for a more efficient detection of melanoma skin cancer. This experimental study shows that fuzzy enhancement integrated with graph-cuts technique achieve a very good segmentation performance on the overall image (i.e. foreground and background) with an average accuracy rate of 97.42%. This study also shows that the background segmentation using fuzzy enhancement combined with graph-cuts technique achieved the good background segmentation with an average specificity rate of 99.07%.