The efficacy of automated "disease/no disease" grading for diabetic retinopathy in a systematic screening programme
- Sam Philip (sam.philip{at}arh.grampian.scot.nhs.uk),
- Alan D Fleming (a.fleming{at}abdn.ac.uk),
- Keith A Goatman (k.goatman{at}biomed.abdn.ac.uk),
- Sofia Fonseca,
- Paul Mcnamee (p.mcnamee{at}abdn.ac.uk),
- Graham S Scotland (g.scotland{at}abdn.ac.uk),
- Gordon J Prescott (gordon.prescott{at}abdn.ac.uk),
- Peter F Sharp (p.sharp{at}biomed.abdn.ac.uk),
- John Alexander Olson (john.olson{at}nhs.net)
- NHS Grampian Retinal Screening Programme, United Kingdom
- Biomedical Physics, University of Aberdeen, United Kingdom
- Biomedical Physics, University of Aberdeen, United Kingdom
- Department of Public Health, University of Aberdeen, United Kingdom
- Health Economics Research Unit, University of Aberdeen, United Kingdom
- Health Economics Research Unit, University of Aberdeen, United Kingdom
- Medical Statistics, Department of Public Health, University of Aberdeen, United Kingdom
- Biomedical Physics, University of Aberdeen, United Kingdom
- NHS Grampian Retinal Screening Programme, United Kingdom
- Published Online First 15 May 2007
Abstract
Aim: To assess the efficacy of automated "disease/no disease" grading for diabetic retinopathy within a systematic screening programme.
Methods: Anonymised images were obtained from consecutive patients attending a regional primary care based diabetic retinopathy screening programme. A training set of 1067 images was used to develop automated grading algorithms. The final software was tested using a separate set of 14406 images from 6722 patients. The sensitivity and specificity of manual and automated systems operating as "disease/no disease" graders (detecting poor quality images and any diabetic retinopathy) were determined relative to a clinical reference standard.
Results: The reference standard classified 8.2% of the patients as having ungradable images (technical failures) and 62.5% as having no retinopathy. Detection of technical failures and any retinopathy was achieved by manual grading with 86.5% sensitivity (95% confidence interval 85.1 to 87.8) and 95.3% specificity (94.6 to 95.9) and by automated grading with 90.5% sensitivity (89.3 to 91.6) and 67.4% specificity (66.0 to 68.8). Manual and automated grading detected 99.1% and 97.9%, respectively, of patients with referable or observable retinopathy/maculopathy. Manual and automated grading detected 95.7% and 99.8%, respectively, of technical failures.
Conclusion: Automated "disease/no disease" grading of diabetic retinopathy could safely reduce the burden of grading in diabetic retinopathy screening programmes.
