Enhancing Skin Cancer Detection with Transfer Learning and Vision Transformers

Early and accurate detection of skin cancer is critical for effective treatment. This research aims to enhance skin cancer multi-class classification using transfer learning and Vision Transformers (ViTs), addressing the challenges of imbalanced medical imaging data. We introduced data augmentation techniques to the HAM10000 dataset to enhance the diversity of the training and implemented 13 pre-trained transfer learning models. These included DenseNet (121, 169, and 201), ResNet (50V2, 101V2, and 152V2), VGG (16 and 19), NasNet (mobile and large), InceptionV3, MobileNetV2, and InceptionResNetV2, as well as two Vision Transformer architectures (ViT and deepViT). After fine-tuning these models, DenseNet121 achieved the highest accuracy of 94%, while deepViT reached 92%, highlighting the effectiveness of these approaches in skin cancer detection. Future work will focus on refining these models, exploring hybrid approaches that combine convolutional neural networks and transformers, and expanding the framework to other cancer types to advance automated diagnostic tools in dermatology.

@article{Ahmad2024,
  title     = {Enhancing Skin Cancer Detection with Transfer Learning and Vision Transformers},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {10},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Istiak Ahmad and Bassma Saleh Alsulami and Fahad Alqurashi},
  doi       = {10.14569/IJACSA.2024.01510104},
  url       = {https://doi.org/10.14569/IJACSA.2024.01510104}
}