AI Detection of Diabetic Retinopathy in Wide-Field Retinal Images
Research type
Research Study
Full title
AI Detection of Diabetic Retinopathy in Wide-Field Retinal Images and Assessment of Non-Mydriatic Image Quality of Wide Field Scanning Confocal Ophthalmoscopes
IRAS ID
275896
Contact name
Peter H Scanlon
Contact email
Sponsor organisation
Gloucestershire Hospitals NHS Foundation Trust
Duration of Study in the UK
1 years, 5 months, 31 days
Research summary
Background
Screening for diabetic eye disease has helped to reduce the rate of blindness in working age people. Diabetic Eye Screening Programmes (DESPs) in England and Wales photograph 60 degrees of the back of the eye (retina) using two overlapping images (2-field) of each eye after dilating the pupil (mydriasis). Scotland takes one 45 degree photograph; mydriasis is used when image quality is poor (staged mydriasis). However, reading the photographs taken in screening is lengthy work and this study will see if we can train computers to read them automatically.
This study is an extension of a previous collaboration between Gloucestershire Hospitals and RetinaScan Ltd in which an algorithm was developed to detect diabetic retinopathy on the standard 45 degree images:
1. The GRRG have worked with the Gloucestershire Diabetic Eye Screening Programme (GDESP) to provide RetinaScan Ltd with a “2016 GDESP real world anonymised image set” containing 120,386 GDESP images and outcome grades that could only be used on the GDESP AI server that they have purchased so that the images would be held within the NHS.
2. GDESP purchased an AI server from Centreprise (Centreprise Renatus AI server).
The current study is looking at the feasibility of extending the use of the algorithm to wider field images.
Wide field scanning confocal ophthalmoscopes (SCO) use either LED light or low-powered laser light to capture a wider-angle image (130 or 200 degrees) without the need for a bright flash. It is thought that most people can be photographed without mydriasis, this may improve attendance.Project Objectives
• To provide wide field retinal images in order to train an algorithm to detect DR.
• To determine the proportion of ungradable images and the need for mydriasis
• To measure time taken to capture and review images
• To determine whether the SCO devices are acceptable to patients and imagers
Study design:
A feasibility study to develop an algorithm for detection of DR on images from wide field SCO cameras and their use in screening
Setting: A single site study in Gloucestershire.
Participants: 400 participants attending primary screening and 100 people attending HEC, 80 with referable DR in one or both eyes prior to treatment and 20 post treatment.
Primary Outcome:
• To develop and train deep learning algorithms to read wide field retinal images in diabetic eye screening
Secondary outcomes:
To assess:
•the proportion of ungradable images by device
• The time taken to capture and review images by device
• The participant and imager perspectives of the devices
Total project duration: 19 monthsREC name
South West - Cornwall & Plymouth Research Ethics Committee
REC reference
20/SW/0009
Date of REC Opinion
3 Feb 2020
REC opinion
Favourable Opinion