Deep learning for ultra-widefield imaging: a scoping review

Abstract

Purpose: This article is a scoping review of published and peer-reviewed articles using deep-learning (DL) applied to ultra-wide eld (UWF) imaging. This study provides an overview of the published uses of DL and UWF imaging for the detection of ophthalmic and systemic diseases, generative image synthesis, quality assessment of images, and segmentation and localization of ophthalmic image features. Methods: A literature search was performed up to August 31st, 20201 using PubMed, Embase, Cochrane Library, and Google Scholar. The inclusion criteria were as follows: (1) Deep Learning, (2) Ultra-Wide eld Imaging. The exclusion criteria were as follows: (1) articles published in any language other than English, (2) articles not peer-reviewed (usually preprints) (3) no full-text availability (4) articles using machine learning algorithms other than deep learning. No study design was excluded from consideration.Results: A total of 36 studies were included. 23 studies discussed ophthalmic disease detection and classi cation, 5 discussed segmentation and localization of UWF images, 3 discussed generative image synthesis, 3 discussed ophthalmic image quality assessment, and 2 discussed detecting systemic diseases via UWFI. Conclusion:The application of DL to UWFI has demonstrated signi cant effectiveness in the diagnosis and detection of ophthalmic diseases including diabetic retinopathy, retinal detachment, and glaucoma. DL has been used with UWFI to also diagnose systemic diseases like Alzheimer's, and also applied in the generation of synthetic ophthalmic images. This scoping review highlights and discusses the current uses of DL with UWFI, and the future of DL applications in this eld.

A Radiomic-Based Machine Learning System to Diagnose Age-Related Macular Degeneration from Ultra-Widefield Fundus Retinography

“…The first component is expected to adapt very well to the standard FRT, since it has already learned all the primitives contained in standard FRT being trained on a wider field of view. However, since standard FRT and UWF-FRT can be translated to each other with DL methods, such as Generative Adversarial Networks [ 40 ], it could be possible, in principle, to develop a system able to generate standard FRT from the dataset used in this work, to be used with a fine-tuning approach; the DL-based macular detector may benefit from a few training epochs in order to specialise on such images. The second component of our method is based on a general approach that extracts the features just from the ROI, while ignoring the surrounding regions.…”

Discriminative-Region Multi-Label Classification of Ultra-Widefield Fundus Images

“…Unlike conventional fundus photographs (FP), which capture only a small portion of the retina with a field of view from 30 to 40 degrees, UFI provides an ultra-wide view of the retina up to 300 degrees. This extended view enables the identification of peripheral diseases that may not be visible in conventional FP [ 1 , 2 ]. However, the low resolution of macular area, frequent artifacts, and distortion of pseudo color images hinder the accurate interpretation of UFI [ 3 ].…”