Research Question:
How effectively can a trained deep learning model (G-RISK) that quantifies glaucomatous damage from color fundus photographs, identify individuals with glaucoma and reveal novel associations with genetic data from a large population-based dataset?
Objectives:
1. To validate G-RISK’s ability to quantify glaucomatous damage using UK Biobank’s extensive fundus photograph dataset by comparing continuous G-RISK scores to ground truth data, such as self-reported glaucoma or related clinical metrics.
2. To correlate G-RISK scores with relevant metadata, including demographic, lifestyle, and clinical factors, to potentially discover unknown associations related to glaucomatous damage.
3. To explore associations between G-RISK scores and genetic data available in the UK Biobank to investigate potential genetic contributions to glaucomatous damage.
Scientific rationale:
Glaucoma is a progressive optic neuropathy and a leading cause of irreversible blindness. Automated quantification of glaucomatous damage can help to lower the large number of undetected cases in the general population. G-RISK is an extensively validated deep learning model that quantifies glaucomatous damage from color fundus photographs.
This project aligns with UK Biobank’s mission to leverage innovative tools for advancing disease understanding and improving health outcomes.
Dissemination of results:
We plan to share the results of our research through a variety of channels to ensure broad accessibility and impact:
1. Academic publication in a peer-reviewed journals focused on ophthalmology, AI in healthcare, and genetics
2. Scientific conferences like the Association for Research in Vision and Ophthalmology (ARVO)
The pre-trained deep learning model will not be publicly shared, nor will it be finetuned on UK Biobank data, complying to the AI Policy.