Development and Evaluation of a Deep Learning System for Screening Retinal Hemorrhage Based on Ultra-Widefield Fundus Images

Abstract

and evaluation of a deep learning system for screening retinal hemorrhage based on ultra-widefield fundus images.

Deep learning from “passive feeding” to “selective eating” of real-world data

“…However, poor-quality images are inevitable in clinical practice due to various factors, such as a dirty camera lens, head/eye movement, eyelid obstruction, operator error, patient noncompliance and obscured optical media 31 , 36 . Therefore, we propose that the systems developed using good-quality images for detecting retinal diseases in real-world settings (e.g., LDRB, retinal detachment, and retinitis pigmentosa) 18 – 28 need to be integrated with the DLIFS to initially discern and filter out poor-quality images, to ensure their optimum performance.…”

“…Currently, in ophthalmology, as ultra-widefield fundus (UWF) imaging becomes a standard-of-care imaging modality for many ocular fundus diseases and a popular tool in screening and telemedicine due to a larger retina area coverage 13 – 17 , an increasing number of studies have developed deep learning-based AI diagnostic systems for automated detection of ocular fundus diseases using UWF images 18 – 28 . To date, all previous UWF image-based AI diagnostic systems have been developed and evaluated using good-quality images alone 18 – 28 . Although the performances of these systems in detecting ocular fundus diseases are ideal in laboratory settings, their performances in real-world settings are unclear because the systems have to make a diagnosis based on images of varying quality.…”

Digital technology, tele-medicine and artificial intelligence in ophthalmology: A global perspective

“…More recently, Li et al developed a DL system to automatically detect the most common sign of DR, retinal haemorrhages, based on 16,827 ultra-widefield fundus (UWF) images (11,339 individuals) from the Chinese Medical Alliance for Artificial Intelligence (CMAAI) ( Li et al, 2020 ). With both sensitivities and specificities over 96% in various settings, this system has significant potential to detect more DR patients, given that the retina view scope of UWF images is five times larger than that of tradition fundus images ( Nagiel et al, 2016 ).…”

Comparison of deep learning systems and cornea specialists in detecting corneal diseases from low-quality images

“…Recently deep learning has attained remarkable performance in disease screening and diagnosis (Cheung et al, 2021;Hosny and Aerts, 2019;Li et al, 2020aLi et al, , 2020bLi et al, , 2020cLi et al, , 2020dMatheny et al, 2019;Zhou et al, 2021). The performance of deep learning is comparable with and even superior to that of human doctors in many clinical image analyses (Li et al, 2021a(Li et al, , 2021b(Li et al, , 2021c(Li et al, , 2021dLi et al, 2020aLi et al, , 2020bLi et al, , 2020cLi et al, , 2020dLi et al, 2019;Ting et al, 2017;Xie et al, 2020;Zhang et al, 2020).…”

“…Recently deep learning has attained remarkable performance in disease screening and diagnosis (Cheung et al, 2021;Hosny and Aerts, 2019;Li et al, 2020aLi et al, , 2020bLi et al, , 2020cLi et al, , 2020dMatheny et al, 2019;Zhou et al, 2021). The performance of deep learning is comparable with and even superior to that of human doctors in many clinical image analyses (Li et al, 2021a(Li et al, , 2021b(Li et al, , 2021c(Li et al, , 2021dLi et al, 2020aLi et al, , 2020bLi et al, , 2020cLi et al, , 2020dLi et al, 2019;Ting et al, 2017;Xie et al, 2020;Zhang et al, 2020). For example, the accuracy of a deep learning system in distinguishing coronavirus pneumonia from computed tomography images reached the level of senior radiologists (87.5% versus 84.5%; p ＞ 0.05) and exceeded the level of junior radiologists (87.5% versus 65.6%; p < .05) (Zhang et al, 2020).…”