SkinNet: A CNN Model for Improved Benign and Malignant Skin Lesion Detection

Skin cancer poses a significant global health challenge, necessitating precise and efficient early detection methods. This study addresses the task of skin cancer detection and classification by employing a hybrid Convolutional Neural Network (CNN) model. The model, comprising two variants named SkinNet1 and SkinNet2, aims to accurately categorize dermoscopy images into benign and melanoma classes, aiding in timely intervention. SkinNet1 and SkinNet2 feature custom architectures, with SkinNet1 relying solely on deep learning techniques and SkinNet2 combining deep learning with Support Vector Machine (SVM). By leveraging deep learning, both models extract intricate features from dermoscopy images, enhancing classification accuracy and facilitating the identification of crucial diagnostic markers. Strong performance is demonstrated through evaluation using an extensive dataset from the International Skin Imaging Collaboration (ISIC) archive. In the validation set, both models achieve an overall accuracy of 88.5% and 90.5%, respectively, which improves in a distinct test set to 90.5% and 92%, highlighting the efficacy of SkinNet2 over SkinNet1. These findings underscore the models' proficiency in discerning subtle visual cues indicative of skin cancer. The convergence of deep learning and medical imaging holds promising potential for advancing healthcare diagnostics.