Enhancing skin disease diagnosis using multilayer perceptron neural network and bat optimization algorithm

Abstract

Skin cancer originates from cells that form the primary components of the skin. These cells grow, divide to form new cells, and replace old ones as they age and die. However, this process can sometimes malfunction, leading to the creation of unnecessary new cells or the failure of old cells to die, resulting in a mass of tissue known as a tumour. In this study, we focused on diagnosing seven types of skin diseases using skin images from the publicly available ISIC dataset. As an innovation, a convolutional neural network architecture called Google Net was employed for optimal feature extraction. Subsequently, the features were classified using a three-layer perceptron network with transfer learning. Before classification, effective features were selected using the Bat Optimization Algorithm in a separate feature selection stage. These optimized features were then fed into the perceptron network for classification. The proposed method achieved an accuracy of 98%, demonstrating a 5% improvement compared to the baseline method.