Enhancing Skin Cancer Classification Using Transformer-Based Deep Neural Networks on Large Datasets

Abstract

The important role of early skin cancer detection for patient prognosis exists despite continuing challenges to achieve effective and accurate classification. Medical imaging analysis through traditional machine learning methods has delivered results although it fails to grasp complex visual features embedded in images. This research investigates transformer-based deep neural networks (DNNs) as a solution for better skin cancer classification from extensive datasets. Utilizing the self-attention fundamental of transformer, we wish to enhance classification precision and broad applicability. The team used publicly accessible skin cancer datasets to compare how transformer models operated versus standard convolutional neural networks (CNNs) models. Our research showed transformer models used alongside extensive diverse datasets outpaced CNN models when performing skin cancer classifications. Studies demonstrate the capabilities of transformer models for medical image analysis while revealing their potential across medical AI solutions.