[Artificial intelligence-based classification for the diagnostics of skin cancer]

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Context:

• Convolutional neural networks (CNNs) demonstrate performance comparable or superior to dermatologists in evaluating pigmented and nonpigmented skin lesions.
• Artificial neural networks analyze images through various layers with graphic filters.
• The first deep learning network for skin lesion analysis received European market approval.

Purpose:

• To evaluate the performance and limitations of deep learning networks in skin lesion assessment.
• To explore the potential of automated total body mapping for future skin cancer screening.

Summary:

• CNNs excel at classifying skin lesions but struggle with rare conditions and image artifacts, potentially leading to misdiagnoses.
• Rare skin entities with limited training data are classified less adequately by current AI models.
• Image artifacts can compromise the diagnostic accuracy of automated systems.

Impact:

• The integration of "man with machine" collaboration is crucial for achieving optimal diagnostic outcomes.
• Automated total body mapping, combining total body photography and lesion assessment, is being explored for future skin cancer screening.
• AI in dermatology offers potential for improved diagnostic accuracy and efficiency in skin cancer detection.