Diagnosis melanoma with artificial intelligence systems: A meta-analysis study and systematic review

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Automatic diagnosis using dermoscopic images with AI and ML is a rapidly advancing research area.
• AI algorithms offer potential for enhanced diagnostic accuracy in clinical settings.
• Melanoma diagnosis is a key application area for AI in dermatology.

Purpose of the Study:

• To synthesize existing studies on the clinical use of AI applications in dermatology.
• To conduct a large-scale meta-analysis of predictive performance for deep learning and hybrid models.
• To summarize the efficacy of AI algorithms in diagnosing skin conditions, particularly melanoma.

Main Methods:

• A comprehensive literature review was performed from January 2006 to May 2024.
• Data were extracted from Web of Science, Scopus, and MEDLINE databases.
• Meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) checklist.

Main Results:

• Evaluated 2722 articles, with data from 78 diagnostic tests across 39 primary studies.
• Overall pooled model AUC was 0.96, sensitivity 0.89, and specificity 0.92.
• Subgroup analysis showed a pooled AUC of 0.98 for HYBRID models.

Conclusions:

• AI and ML algorithms are recommended for extensive use in medicine to aid physicians, especially in melanoma diagnosis.
• HYBRID model algorithms show promising and high performance in disease prediction.
• Findings can guide clinicians in selecting optimal algorithms for melanoma diagnosis.