Skin cancer detection using dual optimization based deep learning network

Abstract

Skin cancer is the most perilous kind of cancer, which is a most important public health problem. Skin cancer can be prevented and treated more effectively if malignant lesions are identified in its early stages. In artificial intelligence, machine learning and deep learning algorithms are used to classify data with high accuracy based on features in the input images. In this work, a novel dual optimization based deep learning network (DODL net) has been proposed for detecting the skin cancer. Initially, the dermoscopic images are gathered from MNIST HAM10000 dataset. The collected images are pre-processed using adaptive median filter and these noise-free images are processed in U-net for segmenting the particular region of the skin lesion. Further, the dual optimization algorithms which is hybridization of Bacterial Foraging Optimization (BFO) and Particle Swarm Optimization (PSO) are utilized for extracting the features from the segmented images. Finally, the Deep Convolutional neural network (CNN) classifies the seven different classes of skin cancer based on the extracted features. The performance of the DODL net has been evaluated using specific parameters such as precision, recall, F1 score and accuracy. The accuracy attained by the proposed DODL net is 98.76% for MNIST HAM10000 dataset.