A Color-Based Approach for Melanoma Skin Cancer Detection

Abstract

Skin cancer cases are continuously arising from the past few years. Broadly skin cancer is of three types: Basal Cell Carcinoma, Squamous Cell Carcinoma, and Melanoma. Among all its types, melanoma is the dangerous form of skin cancer whose treatment is possible only if it is detected in early stages. Early detection of melanoma is really challenging. Therefore, various systems were developed to automate the process of melanoma skin cancer diagnosis. Features used to characterize the disease play a very important role in the diagnosis. It is also very important to find the correct combination of features and the machine learning techniques for classification. Here, a system for the melanoma skin cancer detection is developed by using a MED-NODE dataset of digital images. Raw images from the dataset contain various artifacts so firstly preprocessing is applied to remove these artifacts. Then to extract the region of interest Active Contour segmentation method is used. Various color features were extracted from the segmented part and the system performance is checked by using three classifiers (Naïve Bayes, Decision Tree, and KNN). The system achieves an accuracy of 82.35% on Decision Tree which is greater than other classifiers.