Artificial intelligence models for the automated detection of skin tumors in children: systematic review of diagnostic accuracy and clinical validation

Abstract

Introduction: Pediatric skin tumors are rare, accounting for less than 2% of childhood malignancies, and 20–30% of pediatric melanomas are initially misdiagnosed. Although adult-focused AI models for skin cancer often achieve AUC values above 0.90, their applicability in children is limited by scarce pediatric data and minimal external validation. This review synthesizes current evidence on the diagnostic accuracy and clinical validation of AI models for automated detection of pediatric skin tumors. Methods: We systematically searched PubMed, Embase, Scopus, Web of Science, and the Cochrane Library for studies published from January 2000 through March 2025. Eligible studies evaluated AI-based diagnosis or classification of skin tumors in pediatric populations and reported diagnostic accuracy metrics. Risk of bias was assessed. Results: Nine studies met inclusion criteria, but only two provided primary quantitative pediatric data. One study that incorporated 1,536 pediatric images into model training and tested performance in a pediatric set (n = 674) demonstrated a 7% mean AUC improvement (p = 0.003) and 162 fewer false positives (p = 0.0014). An earlier model trained predominantly on adult images achieved high sensitivity (98%) but low specificity (44%). Three systematic reviews reported diagnostic accuracies of 67–99% for general skin cancer detection, consistently emphasizing the severe shortage of pediatric cases in major image repositories. Additional narrative reviews and one protocol confirmed that fewer than 10% of AI dermatology studies focus on children. Conclusions: AI shows meaningful potential for improving pediatric skin tumor detection, and inclusion of pediatric data clearly enhances performance. However, progress is constrained by limited high-quality datasets and a lack of rigorous validation. Dedicated pediatric image repositories and prospective clinical trials are essential for advancing reliable clinical implementation.

Ähnliche Arbeiten