Skip to content
    Reviews

    AI Incorporating ABCD(E) Criteria for Melanocytic Lesions

    Lopes et alJournal of cancer research and clinical oncology

    A systematic review conducted in accordance with Cochrane and PRISMA guidelines evaluated the diagnostic performance of artificial intelligence (AI) systems incorporating or informed by ABCD(E) criteria (Asymmetry, Border irregularity, Color variegation, Diameter, and Evolution) for melanoma detection. Searching PubMed, LILACS, Latindex, and Cochrane Central Register of Controlled Trials platforms with independent screening and data extraction by two reviewers, the authors assessed 2,735 records and included 21 studies analyzing more than 90,000 images and cases.
    The review found that AI models—specifically convolutional neural networks (CNNs) and deep convolutional neural networks (DCNNs)—demonstrated promising diagnostic performance, frequently matching or outperforming human expert evaluation. For early melanoma differentiation, these systems achieved an Area Under the Curve (AUC) of 90.3%.
    Despite showing significant capability and offering enhanced diagnostic consistency through ABCD(E) integration, critical limitations remain. Patient selection bias was identified across studies, restricting the applicability of findings. In addition, implementation is hindered by ongoing challenges related to data limitations, model generalizability, and algorithmic interpretability.

    Key Takeaways

    • Artificial intelligence systems, notably CNNs and DCNNs incorporating or informed by ABCD(E) criteria, achieved an Area Under the Curve (AUC) of 90.3% for early melanoma differentiation, often matching or exceeding human expert evaluation.
    • Patient selection bias was observed across the 21 evaluated studies (analyzing over 90,000 images/cases), with persistent challenges regarding data limitations, model generalizability, and interpretability limiting real-world applicability.
    Read the original sourceJournal of cancer research and clinical oncology