Sharper than human eyes? A systematic review and meta-analysis of machine learning for retinal detachment detection

Abstract

Retinal detachment (RD) is a sight-threatening condition requiring rapid diagnosis to prevent vision loss. Artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), offers potential for improving diagnostic accuracy in RD, addressing current limitations of manual interpretation. A diagnostic test accuracy (DTA) meta-analysis adhering to PRISMA 2018 guidelines was conducted. Studies utilizing ML for RD detection across imaging modalities were included. Databases (PubMed, Web of Science, Scopus, Cochrane) and manual searches identified 20 studies. Statistical analyses assessed sensitivity, specificity, and area under the curve (AUC), with subgroup analyses by ML technique, imaging modality, validation method, and testing set size. From 69 models analyzed, pooled sensitivity and specificity were 95.7% (95% CI: 94.1–96.9%) and 99.2% (95% CI: 98.7–99.5%), respectively, indicating high diagnostic accuracy. DL models outperformed ML, achieving higher sensitivity (96.3% vs. 92.7%) and specificity (99.4% vs. 96.7%). Models employing fundus imaging exhibited superior performance (sensitivity: 97.3%; specificity: 99.5%). However, significant heterogeneity (I² > 90%) was noted. External validation enhanced specificity but highlighted challenges in generalizability due to biases in patient selection and data quality. ML demonstrates high potential for accurate RD detection, particularly using DL and fundus imaging. Nonetheless, addressing biases, heterogeneity, and external validation is crucial for clinical adoption. Future research should focus on standardization, cost-effectiveness, and multicenter validation to ensure practical ML integration in ophthalmology.