Artificial Intelligence for Skin Cancer Detection: Scoping Review

Area of Science:

• Dermatology
• Medical Informatics
• Computer Science

Background:

• Skin cancer is a prevalent human malignancy.
• Traditional diagnostic methods are time-consuming, expensive, and require expert physicians.
• Artificial intelligence (AI) offers a promising alternative for skin cancer detection and classification.

Purpose of the Study:

• To systematically identify and categorize AI-based technologies for skin cancer detection.
• To evaluate the reliability of AI models by analyzing dataset size and diagnostic class impact on performance metrics.

Main Methods:

• A systematic literature search was conducted across IEEE Xplore, ACM DL, and Ovid MEDLINE databases.
• Studies were selected based on relevance to skin cancer, AI utilization, and diagnostic capabilities, adhering to PRISMA-ScR guidelines.
• Data extraction and synthesis were performed by two independent reviewers, grouping studies by AI technique and evaluation metrics.

Main Results:

• Out of 906 retrieved papers, 53 met the inclusion criteria.
• Shallow AI techniques were employed in 14 studies, while deep AI techniques were used in 39 studies.
• Accuracy was the primary metric in 39 studies, with smaller datasets generally yielding higher reported accuracy.

Conclusions:

• AI models show potential in skin cancer detection, but direct comparison is challenging due to varied metrics and image types.
• Factors like dataset size and number of diagnostic classes significantly influence AI model performance.
• The reliability of high-accuracy AI models is questionable when trained on limited data, necessitating cautious interpretation of results.