Transformer-Based Approach to Melanoma Detection.

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Melanoma is a dangerous skin cancer driven by DNA damage, often from UV exposure.
• Early detection is critical for effective melanoma treatment and survival.
• Current diagnostic methods can be improved with advanced computational tools.

Purpose of the Study:

• To develop and evaluate a Vision Transformer (ViT)-based deep learning architecture for classifying melanoma.
• To assess the model's performance on a public dataset of skin lesions.
• To identify optimal classifier configurations for melanoma diagnosis.

Main Methods:

• Utilized a Vision Transformer (ViT) architecture for image classification.
• Trained and validated the model on the ISIC challenge dataset, a public repository of skin cancer images.
• Analyzed various classifier configurations to determine the most effective setup.

Main Results:

• The best-performing ViT model achieved an accuracy of 0.948.
• Demonstrated high sensitivity (0.928) and specificity (0.967) in identifying melanoma.
• Obtained an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.948.

Conclusions:

• The proposed ViT-based model shows significant promise for accurate melanoma classification.
• This AI approach can aid dermatologists in early and reliable skin cancer diagnosis.
• Further development could enhance clinical decision-making for melanoma detection.