Improved segmentation algorithm using PSO and K-means for basal cell carcinoma classification from skin lesions

Abstract

Skin carcinoma has been sighted as one of the prevalent forms of carcinomas, specifically amongst Caucasian offspring and pale-skinned population. Basal Cell Carcinoma (BCC) is a malevolent skin carcinoma and its classification in earlier stage is a biggest issue. Whilst curable with early detection, only extremely skilled specialists are likely to recognize the disease accurately from skin lesions Dermoscopic images. Since expertise has limited contribution, an automatized system capable of classifying disease can be helpful in saving lives, reducing unnecessary biopsies, and extra costs. On the way to achieve this objective, we proposed a BCC classification model that unifies recent advances in deep learning with Artificial Neural Network (ANN) structure, creating hybrid algorithm of K-means segmentation with Particle Swarm Optimization (PSO) that are capable of segmenting accurate skin lesions region from dermoscopy images, as well as examining the detected region and neighboring tissues for BCC. The proposed system is evaluated using the largest publicly accessible standard skin lesions dataset of dermoscopic images, containing BCC and Non-BCC images. When the evaluation parameters of proposed work are contrasted with a couple of other top-of-the-line techniques, the proposed technique accomplishes superior performance of 97.9% with respect to area under the curve (AUC) in distinguishing BCC from benignant lesions only through the extricated Speeded Up Robust Features (SURF).