Melanoma detection using Particle Swarm Optimized Artificial Neural Network

Abstract

Melanoma, the most dreadful skin cancer with a high mortality rate, is initially diagnosed visually through clinical screening, dermoscopic analysis, biopsy, and histopathological examination. Delays in diagnosis and early treatment can make it more dangerous. Recent developments in image processing techniques aid in detecting melanoma efficiently due to the fine-grained variability in lesions. This paper presents a new classification procedure for analyzing lesion irregularities using Particle Swarm Optimized Artificial Neural Network (PSO-ANN). Color features from the lesion are extracted, and classification is performed using the PSO-ANN classifier. Receiver Operating Characteristics (ROC) obtained from marking false positive and true positive rates play a vital role in analyzing the diagnostic potential of the computer-aided diagnosis system. Classification techniques applied to the ISIC database indicate an area under the curve (AUC) of 0.96853, with a specificity of 90.0%, sensitivity of 94.07%, and accuracy of 93.04%.