Deep Learning in the Diagnosis of Choroidal Melanoma: A Systematic Review and Meta-Analysis

Abstract Purpose: Uveal melanoma (UM), particularly its most common subtype, choroidal melanoma (CM), is the leading intraocular malignancy in adults. Early diagnosis is crucial for improved prognosis, but accurately identifying small or early lesions remains challenging due to limitations in imaging. Artificial intelligence, especially deep learning (DL), shows promise for enhancing diagnosis and risk assessment of choroidal lesions. This study summarized DL models' diagnostic performance, clinical usefulness, limitations, pooled estimates, and research needs for detecting CM.