Multi-Classification of Skin Lesion Images Including Mpox Disease Using Transformer-Based Deep Learning Architectures

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Dermatology

Background:

• Mpox (monkeypox) is a significant global health concern, affecting 110 countries and posing a pandemic risk.
• Traditional Mpox detection methods are slow and expensive, necessitating advanced diagnostic solutions.
• There is a critical need for rapid, accurate, and autonomous methods for Mpox diagnosis using skin lesion images.

Purpose of the Study:

• To develop a multi-class, fast, and reliable autonomous diagnostic model for Mpox and other skin diseases using transformer-based deep learning.
• To evaluate the impact of self-supervised learning, self-distillation, and shifted window techniques on diagnostic accuracy.
• To leverage the Mpox Skin Lesion Dataset (Version 2.0) for training and validation.

Main Methods:

• Utilized transformer-based deep learning architectures, including Vision Transformer (ViT), Masked Autoencoders (MAE), DINO, and SwinTransformer.
• Trained and validated models on the Mpox Skin Lesion Dataset (Version 2.0).
• Investigated the efficacy of self-supervised learning, self-distillation, and shifted window techniques within transformer models.

Main Results:

• The proposed SwinTransformer architecture achieved 93.71% accuracy, outperforming other models.
• SwinTransformer demonstrated an 8% improvement in accuracy compared to the closest competitor.
• ViT, MAE, and DINO achieved accuracies of 93.10%, 84.60%, and 90.40%, respectively.

Conclusions:

• The developed deep learning model successfully diagnoses Mpox and other skin lesions with high accuracy.
• This technology can significantly support clinical decision-making for healthcare professionals.
• Findings offer valuable insights for applying transformer-based models in medical fields with limited image data.