Australian researchers are applying artificial intelligence (AI) technology to optical coherence tomography (OCT) scans in a project designed to provide clinicians with more specific details of an eye’s choroidal tissue.
The study, which was recently published in the journal Nature Scientific Reports, is being undertaken by researchers at the Queensland University of Technology (QUT).
The newly-developed system uses a deep learning network, trained on OCT images, to identify the boundaries of the choroid and the retina. During the 18-month longitudinal study of 101 children with healthy eyes, the AI-powered system produced results that were both more accurate and reliable than standard methods of analysis.
Dr David Alonso-Caneiro, QUT senior research fellow and study lead author, told Insight AI has prompted a significant change in how images are processed.
“The program learns from a large database of images, during an initial training Phase,” he said. “After training, when we presented the program with a new [unseen] image (that has not been used during training) it is able to segment the region of interest automatically and with a very good accuracy.
”The project also represents a slightly different use of AI in eyecare than many of the other research initiatives currently under way. Whereas many focus on automatically diagnosing particular conditions, QUT’s looks at segmentation.
“When a clinician looks at an OCT image they might already be able to objectively pick up some conditions,” Alonso-Caneiro said. “But the whole idea is if you want to pick things up earlier, you want to look at the small changes in the structure.
“I think that’s where this study is crucial. First, we segment the region, and then that way the clinician has the information to track thinning and thickening, and those can be associated with different pathologies.
“Because the choroid supplies the blood and nutrients into the back of the eye, a lot of the pathologies would manifest in changes of the choroid.”
The software has been made freely available to other researchers, which Alonso-Caneiro hopes could translate into patient benefits.
“One of the members of the team is an ophthalmologist, and we have tried to take these concepts into more pathology studies.“The engineering challenge for us is to make this method as robust as possible so we can get excellent accuracy.”
