Revolutionizing dermatopathology using AI in skin diagnostics: scoping review

Area of Science:

• Dermatology and Artificial Intelligence
• Medical Image Analysis
• Computational Pathology

Background:

• Artificial intelligence (AI) is increasingly integrated into skin disease diagnosis and treatment.
• Existing AI models often rely on Convolutional Neural Networks (CNN) and Vision Transformers (ViT).
• Large Language Models (LLMs) are emerging as novel architectures for interactive dermatological analysis.

Purpose of the Study:

• To conduct a scoping review of AI models in skin disease diagnosis and treatment.
• To analyze the evolution of AI architectures, including CNN, ViT, and LLM-based models.
• To assess the diagnostic efficacy of AI models across various skin conditions, including rare diseases.

Main Methods:

• Adherence to PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines.
• Systematic search and selection of 12 articles published between 2017 and 2024.
• Analysis of AI model types, data sources (private clinical data, public datasets like ISIC, MoleMap), and disease coverage.

Main Results:

• CNN and ViT models are prevalent, with a recent increase in LLM-based models like SkinGPT and Gemini.
• AI models demonstrate high efficacy for common skin diseases but show diminished performance for rare or underrepresented conditions.
• Studies utilized diverse data, but most AI models require further clinical validation and regulatory oversight.

Conclusions:

• AI models offer significant potential to improve dermatopathology, enhancing lesion classification and early detection.
• Current AI models are often too generalized and lack robustness for diverse dermatological applications.
• Future development necessitates more robust, clinically validated, and ethically regulated AI tools for healthcare integration.