Robust Melanoma Thickness Prediction via Deep Transfer Learning enhanced by XAI Techniques

AI can help predict melanoma severity: The study used deep learning models to predict Breslow depth (melanoma thickness), a critical measurement for cancer staging and treatment planning, demonstrating that machine learning can assist medical professionals in making more informed diagnostic decisions.

Explainability matters for medical AI: The researchers used explainability techniques (XAI) like feature visualization to show why the AI made its predictions, revealing which visual characteristics in dermoscopy images correlate with melanoma depth—an important step toward building trust in AI-assisted diagnosis.

Model performance improves at higher risk levels: The correlation between the model's predictions and actual melanoma thickness was moderate overall, but improved significantly for thicker (more severe) melanomas, suggesting the AI is most reliable precisely when stakes are highest for patient outcomes.

Data quality and balance are essential: The study combined 1,162 images from multiple sources (ISIC and private datasets) and deliberately balanced the data to ensure robust training, highlighting that responsible AI development requires careful attention to dataset composition and quality.

Transparency about limitations is crucial: The moderate overall correlation suggests the model has limitations that clinicians should understand—responsible deployment would require clear communication that this tool should support, not replace, expert medical judgment.