Research at the intersection of evidence-based ophthalmology and digital health

Project researcher: Dr Arun Thirunavukarasu, Academic Specialised Foundation Programme

Arun’s research interests lie at the intersection between evidence-based ophthalmology and digital health (including artificial intelligence). During his ASFP, he worked with a variety of teams on projects to pursue these interests. This has set him up well for an NIHR Academic Clinical Fellowship in Ophthalmology with the International Centre for Eye Health (London School of Hygiene and Tropical Medicine).

Reviews and meta-research

Supervised by Dr Jasmina Kapetanovic and Prof Robert MacLaren at the Nuffield Laboratory of Ophthalmology, Arun led a systematic review of robot-assisted eye surgery and narrative review on clinical trial endpoints for inherited retinal diseases. He also assisted with a scoping review of functional visual outcomes as clinical trial measures in eye disease. This work led to presentations in Seattle and Rome, and publications in TVST, BMJ Open, and Gene Therapy [forthcoming].

With input from Prof Carl Heneghan at the Centre for Evidence-Based Medicine, Arun led a meta-research study evaluating how investigators allocate treatments to contralateral eyes in randomised-control trials. He found that unbalanced inclusion of one or both eyes from participants is common, with fuller findings presented in Liverpool and published in Eye [forthcoming].

Automation, large language models and continued collaborations

Supervised initially by Dr Bartek Papiez, and subsequently by Dr Le Zhang, Arun worked with the Big Data Institute to improve the performance of automated segmentation of retinal vasculature on fundus photographs. He subsequently also worked on a deep-learning ensemble for diabetic macula oedema recognition (winning 2^nd prize in a MICCAI competition) and a vision-language model for CT scan reporting.

Arun also continued work started while studying in Cambridge to improve identification of glaucoma patients with blindness to increase the proportion that receive social support (supervised by Prof Rupert Bourne at Anglia Ruskin). With a team of University of Cambridge students, he developed a web-application for:

efficient classification of patients with perfect accuracy in a validation study (published in npj Digital Medicine), and
identification of many missed patients in a subsequent cross-sectional study (published in British Journal of Ophthalmology).

This work won prizes from OUCAGS (Bell Session Prize), the UKFPO (Poster Presentation Prize), and the Royal Society of Medicine (Best Paper, Ophthalmology Section).

As principal investigator, Arun led a project using large language models (LLMs) to automate aspects of systematic review, including abstract screening. The team's approach and findings were published in JAMIA. Also, Arun won his first independent grant for this work, from HealthSense.

Arun led a second LLM project supervised by Dr Darren Ting at University of Birmingham, evaluating the clinical reasoning capability of the latest models compared to ophthalmologists at different stages of training. This work was published in PLOS Digital Health, featured by the Financial Times, the BBC, Sky News, Der Spiegel, and other international news media, and awarded the Wallace Foulds Prize by the Royal College of Ophthalmologists.

As a result of previous work on LLMs, Arun was invited to an expert consensus exercise to design a reporting tool for chatbot health advice studies (called ‘CHART’), which will be published in BMJ and other journals later in 2025. The panel also included Prof Gary Collins of the UK EQUATOR Centre, and the team's preliminary scoping review was published in JAMA Network Open. In addition, Arun was invited to give a talk on generative artificial intelligence in Hong Kong, and a longer lecture for the Parliamentary Office of Science and Technology in Westminster.

Other continued collaborations with teams in Cambridge, London, and Singapore led to various other publications and presentations during his Foundation years (see below).

Additional publications and presentations

Thirunavukarasu, A. J. et al. Robot-Assisted Eye Surgery: A Systematic Review of Effectiveness, Safety, and Practicality in Clinical Settings. Translational Vision Science & Technology 13, 20 (2024).
Raji, S., Thirunavukarasu, A. J., Taylor, L. J. & MacLaren, R. E. Functional vision tests as clinical trial outcome measures in ophthalmology: a scoping review. BMJ Open 15, e097970 (2025).
Qin, P., Thirunavukarasu, A. J. & Zhang, L. Deep Learning Ensemble for Predicting Diabetic Macular Edema Onset Using Ultra-Wide Field Color Fundus Image. Preprint at https://doi.org/10.48550/arXiv.2410.06483 (2024).
Thirunavukarasu, A. J. et al. A validated web-application (GFDC) for automatic classification of glaucomatous visual field defects using Hodapp-Parrish-Anderson criteria. npj Digit. Med. 7, 1–4 (2024).
Thirunavukarasu, A. J. et al. Semi-automated screening reveals patients with glaucoma-induced blindness missing out on social support: a cross-sectional study of certificate of visual impairment allocation. Br J Ophthalmol 1–6 (2025) doi:10.1136/bjo-2024-326745.
Sanghera, R. et al. High-performance automated abstract screening with large language model ensembles. JAMIA ocaf050 (2025) doi:10.1093/jamia/ocaf050.
Thirunavukarasu, A. J. et al. Large language models approach expert-level clinical knowledge and reasoning in ophthalmology: A head-to-head cross-sectional study. PLOS Digital Health 3, e0000341 (2024).
Huo, B. et al. Large Language Models for Chatbot Health Advice Studies: A Systematic Review. JAMA Network Open 8, e2457879 (2025).
Thirunavukarasu, A. J. et al. Electron beam-irradiated donor cornea for on-demand lenticule implantation to treat corneal diseases and refractive error. Acta Biomaterialia 169, 334–347 (2023).
Tan, T. F. et al. Artificial intelligence and digital health in global eye health: opportunities and challenges. The Lancet Global Health 11, e1432–e1443 (2023).
Yang, R. et al. Large language models in health care: Development, applications, and challenges. Health Care Science 2, 255–263 (2023).
Ng, F. Y. C. et al. Artificial intelligence education: An evidence-based medicine approach for consumers, translators, and developers. CR Med 4, 101230 (2023).
Thirunavukarasu, A. J. et al. Democratizing Artificial Intelligence Imaging Analysis With Automated Machine Learning: Tutorial. Journal of Medical Internet Research 25, e49949 (2023).
Basu, S., Thirunavukarasu, A. J., Maheshwari, V., Zode, M. & Hassan, R. Burden, determinants and treatment status of metabolic syndrome among older adults in India: a nationally representative, community-based cross-sectional survey. BMJ Public Health 1, (2023).
Li, Y. et al. The next generation of healthcare ecosystem in the metaverse. Biomedical Journal 100679 (2023) doi:10.1016/j.bj.2023.100679.
Thirunavukarasu, A. J. How Can the Clinical Aptitude of AI Assistants Be Assayed? Journal of Medical Internet Research 25, e51603 (2023).
Sienko, A., Thirunavukarasu, A. J., Kuzmich, T. & Allen, L. An Initial Validation of Community-Based Air-Conduction Audiometry in Adults With Simulated Hearing Impairment Using a New Web App, DigiBel: Validation Study. JMIR Form Res 8, e51770 (2024).
Thirunavukarasu, A. J. et al. Nanohydroxyapatite Coating Attenuates Fibrotic and Immune Responses to Promote Keratoprosthesis Biointegration in Advanced Ocular Surface Disorders. ACS Appl. Mater. Interfaces 16, 25892–25908 (2024).
Thirunavukarasu, A. J. et al. Clinical performance of automated machine learning: a systematic review. Ann Acad Med Singap 53, 187–207 (2024).
Thirunavukarasu, A. J. & O’Logbon, J. The potential and perils of generative artificial intelligence in psychiatry and psychology. Nat. Mental Health 1–2 (2024) doi:10.1038/s44220-024-00257-7.
Sinha, A., Thirunavukarasu, A. J., Bonshahi, A. & Brassett, C. Impact of Anatomical Research Projects for Medical Students: A Cross-Sectional Survey of Academic and Professional Skills, Clinical Aspirations and Appreciation of Anatomy. Clinical Anatomy 38, 347–354 (2025).