A robust deep learning framework for multiclass skin cancer classification

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Skin cancer is a major global health issue, with early diagnosis crucial for effective treatment and survival.
• Accurate classification of skin lesions is challenging due to visual similarities between benign and malignant types.

Purpose of the Study:

• To develop an innovative hybrid deep learning model for enhanced skin lesion classification.
• To improve the accuracy and efficiency of early skin cancer diagnosis.

Main Methods:

• A hybrid deep learning model combining ConvNeXtV2 blocks and separable self-attention mechanisms was proposed.
• The model was trained and validated on the ISIC 2019 dataset, utilizing data augmentation and transfer learning.

Main Results:

• The proposed model achieved 93.48% accuracy, 93.24% precision, 90.70% recall, and 91.82% F1-score.
• It outperformed numerous Convolutional Neural Network (CNN) and Vision Transformer (ViT) based models.
• The model has a compact design with 21.92 million parameters, ensuring efficiency.

Conclusions:

• The developed model demonstrates high accuracy and generalizability for diverse skin lesion classification.
• It offers a reliable framework for early and precise skin cancer diagnosis in clinical settings.