Feasibility of an automated AI-based screening tool for diabetic retinopathy at an endocrine outpatient clinic in SA

Simple item page
dc.contributor.advisorSteffen, Jonel
dc.contributor.authorRoux, Margaretha Magdalena
dc.date.accessioned2025-03-28T19:47:49Z
dc.date.available2025-03-28T19:47:49Z
dc.date.issued2024
dc.date.updated2025-03-28T19:45:25Z
dc.description.abstractINTRODUCTION: Diabetic retinopathy (DR) is a worsening global pandemic and a leading cause of blindness. Screening is paramount. In the South African public health sector, screening initiatives have faced significant challenges and leveraging new screening technologies may prove useful. This study aimed at evaluating the feasibility of an autonomous AI-based diagnostic tool in an endocrine outpatient clinic at Groote Schuur Hospital. METHODS: Patients identified as referable DR (moderate NPDR or worse) by autonomous AI screening, as well as patients with ungradable images, were referred to an ophthalmologist. We assessed the time it took to do screening, number of patients requiring dilation, number of ungradable images and their potential causes, referral burden, and number of patients requiring treatment. RESULTS: A total of 62 patients underwent screening, with a median AI screening time of 11.7 minutes. Of these, 55 (88.7%) required referral to ophthalmology. This included 36 patients (58.1%) with referable DR according to AI grading (of which 19 patients (30.6%) had vision-threatening DR) and 19 (30.6%) with ungradable images despite dilatation. Nine patients (14.5%) were lost to follow-up between AI screening and ophthalmology assessment, and 8 patients (12.9%) required treatment for vision-threatening DR according to ophthalmology human grading. Cataracts were the most important cause for ungradable images. . CONCLUSION: This study showed that screening for diabetic retinopathy using autonomous AI is feasible in terms of time. However, the significant burden of referrals and high number of ungradable images may be problematic within a resource-constrained public healthcare system.
dc.identifier.apacitationRoux, M. M. (2024). <i>Feasibility of an automated AI-based screening tool for diabetic retinopathy at an endocrine outpatient clinic in SA</i>. (). University of Cape Town ,Faculty of Health Sciences ,Division of General Surgery. Retrieved from http://hdl.handle.net/11427/41292en_ZA
dc.identifier.chicagocitationRoux, Margaretha Magdalena. <i>"Feasibility of an automated AI-based screening tool for diabetic retinopathy at an endocrine outpatient clinic in SA."</i> ., University of Cape Town ,Faculty of Health Sciences ,Division of General Surgery, 2024. http://hdl.handle.net/11427/41292en_ZA
dc.identifier.citationRoux, M.M. 2024. Feasibility of an automated AI-based screening tool for diabetic retinopathy at an endocrine outpatient clinic in SA. . University of Cape Town ,Faculty of Health Sciences ,Division of General Surgery. http://hdl.handle.net/11427/41292en_ZA
dc.identifier.ris TY - Thesis / Dissertation AU - Roux, Margaretha Magdalena AB - INTRODUCTION: Diabetic retinopathy (DR) is a worsening global pandemic and a leading cause of blindness. Screening is paramount. In the South African public health sector, screening initiatives have faced significant challenges and leveraging new screening technologies may prove useful. This study aimed at evaluating the feasibility of an autonomous AI-based diagnostic tool in an endocrine outpatient clinic at Groote Schuur Hospital. METHODS: Patients identified as referable DR (moderate NPDR or worse) by autonomous AI screening, as well as patients with ungradable images, were referred to an ophthalmologist. We assessed the time it took to do screening, number of patients requiring dilation, number of ungradable images and their potential causes, referral burden, and number of patients requiring treatment. RESULTS: A total of 62 patients underwent screening, with a median AI screening time of 11.7 minutes. Of these, 55 (88.7%) required referral to ophthalmology. This included 36 patients (58.1%) with referable DR according to AI grading (of which 19 patients (30.6%) had vision-threatening DR) and 19 (30.6%) with ungradable images despite dilatation. Nine patients (14.5%) were lost to follow-up between AI screening and ophthalmology assessment, and 8 patients (12.9%) required treatment for vision-threatening DR according to ophthalmology human grading. Cataracts were the most important cause for ungradable images. . CONCLUSION: This study showed that screening for diabetic retinopathy using autonomous AI is feasible in terms of time. However, the significant burden of referrals and high number of ungradable images may be problematic within a resource-constrained public healthcare system. DA - 2024 DB - OpenUCT DP - University of Cape Town KW - Diabetic retinopathy LK - https://open.uct.ac.za PB - University of Cape Town PY - 2024 T1 - Feasibility of an automated AI-based screening tool for diabetic retinopathy at an endocrine outpatient clinic in SA TI - Feasibility of an automated AI-based screening tool for diabetic retinopathy at an endocrine outpatient clinic in SA UR - http://hdl.handle.net/11427/41292 ER - en_ZA
dc.identifier.urihttp://hdl.handle.net/11427/41292
dc.identifier.vancouvercitationRoux MM. Feasibility of an automated AI-based screening tool for diabetic retinopathy at an endocrine outpatient clinic in SA. []. University of Cape Town ,Faculty of Health Sciences ,Division of General Surgery, 2024 [cited yyyy month dd]. Available from: http://hdl.handle.net/11427/41292en_ZA
dc.language.isoen
dc.language.rfc3066eng
dc.publisher.departmentDivision of General Surgery
dc.publisher.facultyFaculty of Health Sciences
dc.publisher.institutionUniversity of Cape Town
dc.subjectDiabetic retinopathy
dc.titleFeasibility of an automated AI-based screening tool for diabetic retinopathy at an endocrine outpatient clinic in SA
dc.typeThesis / Dissertation
dc.type.qualificationlevelMasters
dc.type.qualificationlevelMMed
Files
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
1.72 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
Masters