Deep Learning for Skin Melanocytic Tumors in Whole-Slide Images: A Systematic Review

Area of Science:

• Digital pathology
• Artificial intelligence in medicine
• Computational pathology

Background:

• Melanoma histological interpretation presents challenges due to interobserver variability among dermatopathologists.
• Artificial intelligence (AI) is emerging as a valuable support tool in clinical pathology workflows.
• Whole-slide imaging (WSI) enables advanced computational analysis of tissue samples.

Purpose of the Study:

• To systematically review studies utilizing deep learning techniques for automatic image analysis of melanocytic tumors on WSI.
• To analyze the application and performance of AI in dermatopathology, focusing on melanoma.
• To identify parameters and conditions crucial for effective AI implementation in real-world pathology settings.

Main Methods:

• Systematic literature search conducted across Embase, Pubmed, Web of Science, and Virtual Health Library.
• Adherence to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist.
• Inclusion of 28 studies published between 2015 and July 2022, with a focus on AI methodologies and clinical objectives.

Main Results:

• Studies were categorized into four groups: pathologist vs. deep learning models (10 studies), diagnostic prediction (7 studies), prognosis (5 studies), and histological features (6 studies).
• Analysis focused on the diverse clinical objectives and AI methods employed in the reviewed literature.
• Identified common parameters and conditions influencing AI performance in pathology.

Conclusions:

• Deep learning on WSI shows promise for improving diagnostic accuracy and efficiency in melanoma analysis.
• Further research is needed to optimize AI algorithms and validate their performance in diverse clinical scenarios.
• Understanding the factors influencing AI performance is essential for successful integration into routine pathology practice.