Chia, Mark Ashley; (2024) Diabetic retinopathy in Indigenous Australians: applications of artificial intelligence and retinal image analysis. Doctoral thesis (Ph.D), UCL (Unviversity College London).

Text Chia_10198533_Thesis_sigs_removed.pdf Access restricted to UCL open access staff until 1 November 2025. Download (20MB)

Abstract

Indigenous Australians are an underserved ethnic group who suffer disproportionately from the burden of diabetic blindness. This thesis aims to enhance the care of diabetic retinopathy (DR) for Indigenous Australians by leveraging novel technology, including artificial intelligence (AI) and advanced image analysis. This aim was pursued through five distinct studies. First, we performed a systematic review of DR prevalence in Indigenous and non-Indigenous Australians, with a view to exploring variation in methodology, characteristics, and prevalence outcomes. We found that Indigenous patients have a higher DR prevalence, but only after adjusting for age and study quality in meta-regression models. This finding highlights the need for ongoing work to address this disparity. Towards this end, we performed a validation study of an AI system for the detection of DR in Indigenous Australians. We found that the model performed well, with higher sensitivity and similar specificity compared to a retinal specialist. Building on this, we explored alternative strategies for creating models that generalise well to diverse populations. We developed an open-source retinal foundation model trained on 1.6 million images using self-supervised learning. We showed that the model could be efficiently fine-tuned for the tasks of DR severity grading and prediction of myocardial infarction. We also noted improved performance amongst ethnic subgroups, illustrating the potential for robust adaptation to an Indigenous population. Finally, we pursued work towards developing an enhanced DR grading system by using segmentation models to extract metrics from fundus photography, optical coherence tomography (OCT), and OCT angiography. We discovered associations between DR severity and several quantitative metrics, which are potential candidates for a modern DR grading system capable of individual risk-stratification. Enhancing healthcare for Indigenous Australians stands as a critical priority in Australia. The adoption of new technology to address resource limitations represents a promising pathway toward this goal.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item