Etudiant : SAFIR EL-HOUSSAINE

Filière : Master Informatique Décisionnelle et Vision Intelligente (MIDVI)

Encadrant : Pr. SABRI ABDELOUAHED

Annèe : 2023

Résumé : Accurate diagnosis of skin lesions plays a critical role in early detection and effective treat- ment of various dermatological conditions. However, it remains a challenging task due to the complexity and visual variability of skin lesion images. In this study, we address this challenge by exploring the application of Vision Transformer (ViT) for skin lesion classification, aiming to assist dermatologists in improving diagnostic accuracy. Skin lesion classification traditionally relies on methods like Convolutional Neural Networks (CNNs). While CNNs have shown promise, they may struggle to effectively capture complex spatial relationships present in skin lesion images. To overcome this limitation, we investigate the potential of ViT, a novel architecture based on the self-attention mechanism. The implications of this research are significant, as accurate and efficient skin lesion clas- sification can aid dermatologists in making informed decisions and improve patient outcomes. Our findings highlight the potential of ViT to surpass the limitations of traditional approaches and provide a robust framework for skin lesion analysis. Future work could focus on fine-tuning strategies, larger datasets, and interpretability techniques to further enhance the model’s per- formance and provide insights into the decision-making process.