Vitreoretinal disease detection using artificial intelligence: a systematic review and meta-analysis

Comprehensive Summary

This systematic review with meta-analysis of 195 studies aimed to assess the diagnostic capability of AI systems in the detection of vitreoretinal diseases using various retinal imaging modalities. The studies comprised a broad spectrum of diseases such as diabetic retinopathy, age-related macular degeneration, retinal vascular diseases, macular edema, retinal detachment, and myopic retinopathy. The majority of studies used deep learning-based models such as convolutional neural networks on fundus photographs and optical coherence tomography images. The results showed good diagnostic accuracy and sensitivity of AI systems in detecting most of these diseases, with the best results of convolutional neural network-based systems over traditional machine learning-based systems. However, a large degree of heterogeneity was apparent in all studies. In spite of such good diagnostic metrics, limitations such as possible bias in study populations and lack of external validation existed. Based on these results, AI systems have a good potential in becoming useful diagnostic tools for the detection of vitreoretinal diseases.

Outcomes and Implications

With the integration of AI, it can be seen that vitreoretinal diseases may present a great opportunity to increase and streamline existing screening platforms, leading to early detection and timely intervention of vitreoretinal diseases, sight-threatening diseases in particular. As part of this effort, AI can potentially reduce clinicians' burden, streamline existing service protocols, and minimize clinicians' inter-observer variability in image interpretation, promoting standardized clinical practices. Furthermore, AI-based telemedicine platforms may also offer new solutions to vitreoretinal care, catering to populations in need and in underserved or rural environments. Nevertheless, before AI-based vitreoretinal diagnostics systems enter an external validation paradigm, several issues need to be taken into consideration by clinicians and researchers alike, including issues of generality, variability, regulatory, and medicolegal issues.