Convolutional Neural Network with Coordinate Attention Module for the Detection of Skin Cancer

Abstract

Skin cancer is one of the most common types of cancer globally, by increasing occurrence rates each year. It is a predominant kind of cancer which rises from the uncontrolled growth of abnormal skin cells due to genetic mutations including various factors such as UV radiation, genetics and other factors. The death rate is decreased when skin cancer is detected at early stages. Therefore, this paper proposed a Convolutional Neural Network (CNN) with Coordinate Attention Module (CAM) for early detection of skin cancer. The data augmentation is utilized in this experiment for data pre-processing and fed into the Gray Level Co-occurrence Matrix (GLCM) based feature extraction technique. The Harris Hawk Optimization (HHO) is utilized for selecting features that have faster convergence and strong capability in local optima. The selected features are given as input to CNN with the CAM approach. This model is estimated on ISIC-2019 and ISIC-2020 datasets and attains better results using accuracy, precision, recall, specificity, and F1-score. The obtained result shows that the proposed model achieves better accuracy of 98.77% on the ISIC-2019 dataset and 99.51% on the ISIC-2020 dataset which ensures accurate detection compared to other existing methods like Inception-ResNet and Residual Deep Convolutional Neural Network (RDCNN).