SkinViT: A transformer based method for Melanoma and Nonmelanoma classification

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Skin cancer is a significant global health issue requiring early diagnosis for effective treatment.
• Automated classification of Melanoma and Nonmelanoma is challenging due to visual similarities and intra-class variations.
• Existing Convolutional Neural Networks (CNNs) struggle to capture global contextual information, potentially missing crucial diagnostic features.

Purpose of the Study:

• To develop an automated system for classifying Melanoma and Nonmelanoma (including Basal Cell Carcinoma and Squamous Cell Carcinoma).
• To enhance the accuracy of skin cancer detection by integrating an outlook attention mechanism to capture both local and global features.
• To assist dermatologists in the timely diagnosis and treatment of skin cancer patients.

Main Methods:

• Proposed a novel architecture, SkinViT, combining outlooker blocks, transformer blocks, and MLP head blocks.
• Utilized an outlook attention mechanism to improve feature representation by highlighting important features and suppressing noise.
• Evaluated the model on three datasets (ISIC2019, online databases, combined) using metrics like accuracy, recall, precision, and F1 score.

Main Results:

• SkinViT achieved high accuracy: 0.9109 on Dataset1, 0.8611 on Dataset2, and 0.8911 on Dataset3.
• The proposed method demonstrated superior performance compared to state-of-the-art models in skin cancer classification.
• SkinViT effectively captured both fine-grained and global contextual information for improved classification accuracy.

Conclusions:

• The SkinViT architecture offers a promising approach for accurate and efficient classification of Melanoma and Nonmelanoma.
• This AI-driven tool can significantly aid dermatologists in early and precise skin cancer diagnosis.
• The study highlights the potential of attention mechanisms in medical image analysis for improved diagnostic outcomes.