Optimization-Based Feature Selection in Deep Learning Methods for Monkeypox Skin Lesion Detection

Abstract

The monkeypox virus, a member of the orthopoxvirus family, is the infectious illness that causes monkeypox. Monkeypox has recently become more common than in the past, and people in many places have been infected with the virus of this disease. Because monkeypox is a rare and potentially dangerous disease, public health officials are taking control measures to prevent its spread. Monkeypox skin lesions in skin images and chickenpox and measles diseases are difficult to distinguish from each other. Considerable progress has been achieved in detecting skin lesions recently from dermoscopy images using Convolutional Neural Networks (CNN). This study performed feature extraction from skin lesion images using pre-trained CNN models DenseNet-201, ResNet-101, and DarkNet-53 models. The synthetic minority high sampling technique was used to correct the imbalance in the dataset, and a hybrid approach combining binary particle swarm optimization (BPSO) and binary grey wolf optimization (BGWO) was used to select the features. These obtained features were classified by Support Vector Machines. While the accuracy rate for the detection of monkeypox was 94.93 % using pre-trained architectures in the experiments, it was observed that the accuracy rate of monkeypox disease increased to 99.71 % when the balance was adjusted and the feature selection was made with BPSO and BGWO optimization. © 2023 IEEE.