GL-MSFN: a global-local and multi-level attention fusion network for dermoscopic image classification

Key Points

• Dermoscopic image classification accuracy improved, demonstrating the efficacy of a multi-level attention fusion model.
• Accuracy attained a notable 92.5% on test datasets, indicating substantial integration of global-local features.
• Analysis utilizing a novel neural network architecture combines local attentions with global cues for feature extraction.
• Enhanced performance may enable better skin lesion diagnostics; external validation needed for broader applicability.