Feature Fusion with Attention Mechanism for Skin Cancer Classification

Skin cancer is a common and potentially fatal condition that results from the unnatural growth of skin cells. A delayed diagnosis of skin cancer can have negative effects and make treatment more difficult. Deep learning-based computer vision-based skin cancer classification has recently demonstrated promising outcomes in terms of early detection and precise diagnosis. This study presents a method for automatic skin cancer detection using Convolutional Neural Networks (CNNs) and transfer learning. The suggested model combines DenseNet121 and Xception, two cutting-edge pretrained CNNs, to extract useful features from skin lesion images. To prioritize crucial features and contextual information, these features are then combined and passed through an attention block. The model exhibits more than 97% accuracy in categorizing eight types of skin cancers on the ISIC dataset while consuming significantly less training resources. The study highlights the advantages of transfer learning in medical image analysis and shows the potential of deep learning-based skin cancer classification for early detection. The results provide useful information for the creation of effective and trustworthy computer-aided dermatology diagnostic systems.