Identification of Melanoma from Nevus Images

Abstract

Melanoma is one of the fatal type skin cancers. The mortality rate is more severe than the present skin related diseases. Humans are prone to various diseases and are easily infected by them due to their adaptive and constrained lifestyle and melanoma is not an exception. Melanoma spreads faster and is less responsive to treatment in its later stages. Thus, harnessing the disease rate becomes more enigmatic and initial diagnosis is the need. Melanoma and nevus are akin and show analogous symptoms and traits. In order to overcome this, we use the technique of image processing to segregate Melanoma and nevus. The input image is processed in such way that the noise in the image (skin lesion) is removed using median filter and is segmented using improved K-means clustering. From the lesion necessary textural and chromatic features are extracted and a unique feature vector is formulated. Melanoma and Nevus are segregated using both Adaptive Neuro-Fuzzy inference System (ANFIS) and Feed-Forward Neural Network (FFNN). The skin images from DERMIS dataset is used in this work and it has 1023 skin images including 104 melanoma and 917 nevus images. Our proposed methodology provides efficient results having 97.3% and 96.8% accuracy for ANFIS and FFNN classifiers.