Integration of retinal photography and artificial intelligence to build opportunistic screening services in primary care settings
Professor Mingguang HeFull Bio
The University of Melbourne
Centre for Eye Research Australia, Royal Victorian Eye and Ear Hospital, St Vincent’s Hospital Melbourne, University of Melbourne
Project summaryDownload PDF
If we could do routine photography of people’s eyes we could screen for these problems but unfortunately examining these images would require a huge additional trained workforce.
This project removes the need to train additional professionals to examine retinal photographs by using an artificial intelligence system to do the work for us. A technique called deep learning allows photographs to be rapidly examined so diseases can be predicted early. Before the system can be utilised however it needs to have 100,000 already labelled images to learn from. Some of these images can be sourced from the Google DeepMind Project but this does not cover all diseases – for example glaucoma. More data is needed.
This project will use a web-based system that incorporates an artificial intelligence (AI)-based automatic image assessment algorithm with proven efficiency and accuracy. Some 85% of the Australian population will visit a GP at least once every 12 months – routine simple eye imaging will be used at GP visits and those images will be automatically assessed by the AI system. This will be a real-world study to assess the impact, feasibility, and cost-effectiveness of this model in a randomised trial.
This project has increased the effectiveness with which we can detect eye diseases before symptoms become noticeable. To enable successful translation of the research into clinical practice, this project study and project team have:
1) Received Therapeutic Goods Administration (TGA) approval for the software as a decision support system
2) Determined that patients are amenable to routine scanning using the AI system
3) Developed a do-it-yourself eye screening booth (prototype) which requires very little human intervention to reduce staff time spent on screenings
4) Commenced further trials of the system across Australia including within remote Aboriginal communities
1. Li, Z., Keel, S., Liu, C., He, Y., Meng, W., Scheetz, J., … & Taylor, H. (2018). An automated grading system for detection of vision-threatening referable diabetic retinopathy on the basis of color fundus photographs. Diabetes care, 41(12), 2509-2516.
2. Keel, S., Wu, J., Lee, P. Y., Scheetz, J., & He, M. (2019). Visualizing deep learning models for the detection of referable diabetic retinopathy and glaucoma. JAMA ophthalmology, 137(3), 288-292.
3. Keel, S., Lee, P. Y., Scheetz, J., Li, Z., Kotowicz, M. A., MacIsaac, R. J., & He, M. (2018). Feasibility and patient acceptability of a novel artificial intelligence-based screening model for diabetic retinopathy at endocrinology outpatient services: a pilot study. Scientific reports, 8(1), 1-6.
4. Tan, Z., Scheetz, J., & He, M. (2019). Artificial intelligence in ophthalmology: accuracy, challenges, and clinical application.
5. Keel, S., Li, Z., Scheetz, J., Robman, L., Phung, J., Makeyeva, G., … & Guymer, R. (2019). Development and validation of a deep‐learning algorithm for the detection of neovascular age‐related macular degeneration from colour fundus photographs. Clinical & Experimental Ophthalmology, 47(8), 1009-1018.
6. Li, Z., He, Y., Keel, S., Meng, W., Chang, R. T., & He, M. (2018). Efficacy of a deep learning system for detecting glaucomatous optic neuropathy based on color fundus photographs. Ophthalmology, 125(8), 1199-1206.