Skin Cancer Detection using Constrastive Learning and Swin Transformer

  • Unique Paper ID: 177694
  • Volume: 11
  • Issue: 12
  • PageNo: 1236-1243
  • Abstract:
  • Early detection and classification of skin cancer are important for effective treatment and improved patient outcomes. This paper presents a novel approach to automated skin cancer classification using a Swin Transformer architecture enhanced with supervised contrastive learning. We address the challenges of class imbalance in skin lesion datasets through weighted random sampling and implement a multi-component loss function combining focal loss, label smoothing, and super- vised contrastive learning. Using the ISIC (International Skin Imaging Collaboration) dataset containing nine classes of skin lesions, our model achieves robust generalization with significant improvement in classification accuracy. The implementation of exponential moving average (EMA) and advanced augmentation techniques further enhances model stability. Our experimental results demonstrate the effectiveness of the proposed approach compared to conventional convolutional neural network methods, offering promising potential for clinical application in dermatological diagnosis.

Cite This Article

  • ISSN: 2349-6002
  • Volume: 11
  • Issue: 12
  • PageNo: 1236-1243

Skin Cancer Detection using Constrastive Learning and Swin Transformer

Related Articles