The future of glaucoma diagnosis

Applying AI to glaucoma diagnosis and management

Dermot McGrath

Posted: Thursday, July 30, 2020

Artificial intelligence (AI) has the potential to revolutionise the screening, diagnosis and classification of glaucoma in the near future, according to Xiulan Zhang MD, PhD.

“The future belongs to AI. Thanks to advanced algorithms, AI will sharply improve glaucoma diagnosis. We are making rapid progress and research is already well advanced on the development of an AI-based diagnostic and decision-making platform for glaucoma,” she told delegates attending the 37th Congress of the ESCRS in Paris.

Glaucoma diagnosis depends primarily on visual field testing and OCT evaluations, said Prof Zhang, Director of the Clinical Research Centre at Zhongshan Ophthalmic Centre (ZOC) in Zhongshan, China. She noted that it is particularly challenging for AI systems, with population differences in fundus pigmentation, disc size and cup-to-disc ratio introducing potential sources of bias in the algorithms.

Much of her team’s research in recent years has been devoted to overcoming some of these challenges by building robust and diverse datasets based on thousands of visual field tests, OCT scans and fundus photographs.

The first step was to successfully develop an algorithm to differentiate glaucomatous visual field from non-glaucomatous visual field, and verify the efficacy of the algorithm in visual field classification in a multi-centre diagnostic trial after collecting more than 10,000 sample visual field tests from seven different eye centres in China.

“We assessed the performance of the algorithm with different input data in the primary validation set. Performance was still better than the glaucoma experts even though the machine-made diagnosis based on only received visual field reports,” said Prof Zhang.

Based on the study, in 2019 the team released a mobile phone and computer application called iGlaucoma 1.0 for visual field interpretation. After inputting Humphrey visual field reports or capturing printed visual field reports, the app can quickly and accurately output diagnosis, with an accuracy rate as high as 87.6%, said Prof Zhang.

“We are conducting a multi-centre clinical trial at the moment to assess the diagnostic efficacy of the algorithm in differentiation of visual field at 19 different eye centres in China. The ultimate goal is the construction of a glaucoma diagnostic and screening platform, iGlaucoma 2.0, which will combine visual field, OCT and fundus photos, and which will hopefully be available soon,”she said.

The idea is that the app can be used by non-glaucoma ophthalmologists and doctors in rural hospitals, to help them to interpret the visual field reports and improve the diagnosis of glaucoma.

Research is also continuing into algorithm development to differentiate angle width and locate the scleral spur in glaucoma patients using OCT scans. Prof Zhang said that the goal was to eventually incorporate this algorithm into a more advanced iGlaucoma 3.0 app capable of interpreting anterior and posterior segment imaging data.

Xiulan Zhang: