Use this URL to cite or link to this record in EThOS:
Title: The Role of Automation in a Scottish Diabetic Retinopathy Screening Programme
Author: Philip, Sam
ISNI:       0000 0001 3487 0870
Awarding Body: University of Aberdeen
Current Institution: University of Aberdeen
Date of Award: 2007
Availability of Full Text:
Access from EThOS:
Aims: 1. To assess the efficiency of the three level manual grading recommended by the Health Technology Board of Scotland. 2. To develop and validate a fully automated level one grading system for diabetic retinopathy screening a. To develop an image grading database and use it to train and validate the clinical 1'. research fellow as a clinicalreference standard. b. To improve the existing image quality assessment criteria and to develop an automated image quality assessment system for retinal images that can reliably detect technical failures. c. To incorporate the image quality assessment software with mIcro-aneurysm detection to develop a fully automated level one grading system for diabetic retinopathy. d. To validate the automated level one retinopathy detection system and compare its efficacy with the manual level-one grading system. ,. 3. To estimate the cost effectiveness of automated level one grading if incorporated within the Scottish national screening programme. Methods: Anonymized images, grading information and outcomes were obtained from the 'Grampian Diabetes Retinal Screening Programme'. A training set of 1067 images was used to develop the automated grading algorithms, while the test set used to assess the performance of the final software consisted of 14,406 images from 6,722 patients. All images were graded for image quality and retinopathy by the reference standard. The performance of the manual graders and the software as graders were compared against the reference standard. An economic evaluation was performed to assess the impact of introducing automated level-one grading in Scotland. Results: Of the 6722 patients, 3725 (55%) were male. The mean age of patients screened was 63 years (±15 years). The reference standard classified 8.2% of the patients as technical failures and 62.5% as having no retinopathy. Ofthe 2523 cases identified as abnormal by the reference standard (retinopathy or technical failures), a greater proportion (341 (12.5%)) were missed by manual grading as opposed to automated (240 (9.5%)). Manual graders identified 99.1% of the patients with referable retinopathy as abnormal compared to 97.9% by the automated grading. Automated grading performed better at identifying quality failures (99.8% vs 95.7%)). Automated grading reduced the overall manual grading workload by 60%. The clear error.,.rate for automated grading was 1 in 1680 compared to a clear error rate of 1 in 448 for manual grading. Conclusions: The three level manual grading recommended by the Health Technology Board of Scotland was found to effective in triaging patients needing review by the level-three grader and minimizing the number ofun-necessary referrals to ophthalmology. An image grading database incorporating an image viewer with facilities for quality control was developed. I was trained as the clinical reference standard and showed excellent agreement with the lead clinician (1(=0.91) for grading outcomes. I developed and valid~ted an image quality assessment criteria and this was used successfully to automate image quality assessment. The automated image quality system was excellent at identifying poor quality images. This was incorporated into a fully automated level one grading system for diabetic retinopathy. The automated grading software had a comparable efficacy to manual level-one grader for detecting referable cases and was found to be more cost-effective. Hence, our automated grading software can perform safely as a level-one grader in a large screening programme using high resolution (2160 x 1440) colour dig~tal images.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (M.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available