• on 30/10/2020

Dr Carol Cheung was interviewed by Omnihealth Practice for her artificial intelligence screening system for glaucoma

Our Assistant Professor, Dr Carol Cheung was interviewed by Omnihealth Practice for her artificial intelligence screening system for glaucoma. Dr Emma Ran is also a core member of this project.

Q : In a publication by your research group it was mentioned that the number of people with glaucoma worldwide will increase to 111.8 million in 2040, disproportionally affecting people residing in Asia and Africa. What is the expected glaucoma-related burden in Hong Kong? What are the measures taken to reduce the prevalence or complications of the disease?

Dr Carol Cheung :

Glaucoma is a complicated disease in which damage to the optic nerve leads to progressive, irreversible vision loss. It is estimated that for those aged 40 years or above, 1 in 38 has glaucoma. In Hong Kong, about a quarter of permanent blindness in Hong Kong is due to glaucoma.

Early glaucoma usually starts from peripheral visual loss, which may NOT be noticeable by the patient. Without prompt therapy, glaucoma may result in complete and irreversible blindness. Therefore, early detection of glaucoma is crucial for timely treatment and minimizing irreversible vision loss. However, population screening for glaucoma is not cost-effective nor practical, as the overall glaucoma prevalence is low and glaucoma detection requires multiple expensive testing including a subjective examination of optic nerve head that experts often do not agree among themselves, especially for identification of early asymptomatic cases. At present, ‘glaucoma screening’ is largely conducted through opportunistic case identification in eye clinics, or amongst people having regular eye check-ups.


Q : In a population-based sample of Asian adults with diabetes and hypertension, treated but poorly controlled as well as untreated hypertension were significantly associated with any-diabetic retinopathy. Would the similar factors affect glaucomatous optic neuropathy? If so, how would these factors contribute to the disease?

Dr Emma Ran :

Glaucoma is a multifactorial disease. Some studies have reported that diabetes mellitus and fluctuation of blood pressure are associated with primary open angle glaucoma. Besides, uncontrolled diabetes mellitus can also cause neovascular glaucoma if there are new blood vessels on the iris. The exact mechanisms are still unclear. Currently, different researchers are investigating how these systemic factors contribute to glaucoma.


Q : Recently you have introduced a three-dimensional deep-learning system based on spectral-domain optical coherence tomography data that can be used to detect glaucomatous optic neuropathy, how much specificity and sensitivity do you expect from this test in the clinical settings?

Dr Carol Cheung :

Optical coherence tomography (OCT) is a non-contact and non-invasive optical technology that imaging the optic nerve head in three-dimensional (3D) view within few seconds. OCT has been more and more widely used in eye clinics for identifying structural glaucomatous damage by ophthalmologists. However, similar to other medical imaging, expertise in glaucoma are required to interpret the OCT results.  In our recent reported study, we developed a new 3D artificial intelligence (AI) deep learning algorithm, and let the AI to learn the glaucoma related features from several thousands of 3D OCT data labelled by glaucoma specialists.  We tested our algorithm both internally and externally in unseen datasets. We found that the algorithm achieved a sensitivity of 89% and specificity of 96% in internal testing. In the external testing with three unseen datasets, the algorithm achieved sensitivities of 78-90% and specificities of 79-86%. This testing was based on retrospective datasets.  We are currently conducting a prospective validation study to further test and refine our AI algorithm for real-world clinical setting and we expect that both the sensitivity and specificity will be above 85%.


Q : In addition to spectral-domain optical coherence tomography, what are the other tests used in detecting glaucomatous optic neuropathy? Among these which test/tests have the highest validity?

Dr Emma Ran :

Other imaging modalities, such as fundus photography, scanning laser polarimetry, and confocal scanning laser ophthalmoscope, are used for glaucomatous optic neuropathy detection. Currently, in clinics, OCT and fundus photography are the most commonly used. We cannot say which tests have highest validity as they all have cons and pros. For example, fundus photography offers qualitative assessment, and is able to storage photos for future comparison. However, the interpretation is subjective and the agreement among experts for the classification of glaucoma based on fundus photographs are not high enough. Comparing with it, OCT offers objective and quantitative measurement, and has higher validity.


Q : Currently, there is a high demand for artificial intelligence-based models for glaucoma screening. How would you expect to utilize these models in clinically to reduce the burden of glaucomatous optic neuropathy?

Dr Carol Cheung :

As I mentioned, conventional approach for population screening for glaucoma is not cost-effective nor practical which currently rely on experienced glaucoma specialists or highly trained assessors to do. AI based screening has huge potential to facilitate glaucoma screening in a more cost-effective way which is more rapid, automated and does not require a large number of experienced trained personnel on site, allowing effective stratification of at-risk patients for further investigations and appropriate referral for management.


Q : The three-dimensional deep-learning system introduced in your study identified that most of the eyes with pre-perimetric glaucoma had glaucomatous optic neuropathy, which might require further examination by glaucoma specialists. Does it mean the proposed three-dimensional deep learning system can be used to identify glaucomatous optic neuropathy early and prevent it?

Dr Carol Cheung :

Yes, we aim to use the proposed AI algorithm to screen for glaucoma, particularly in its early stage to prevent any irreversible vision loss and blindness. By using the proposed AI algorithm, screening positive individuals will then be advised to perform further examination such as visual field test and see glaucoma specialists for confirmation of diseases and disease management.


Q : Person- and area-level socioeconomic status have been associated with the diabetic retinopathy. How these socioeconomical status would affect the glaucomatous optic neuropathy?

Dr Emma Ran :

Studies showed the socioeconomic deprivation is associated with greater severity of glaucoma at presentation to ophthalmologists. As glaucoma is usually asymptomatic, most of patients with glaucoma are diagnosed via opportunistic findings. In more developed area, people are more aware of having regular eye check-ups, and thus are more likely to be detected early. We developed this AI based system, with the aim to ultimately help with people in rural areas or settings with limited resources to better access to glaucoma screening. Of course, further researches are still needed, such as prospective validation of the performance, and incremental cost-effectiveness analysis.


Q : How would you expect to use the three-dimensional deep-learning system using spectral-domain optical coherence tomography data in guiding the treatment of glaucomatous optic neuropathy and enhance the quality of life of patients?

Dr Carol Cheung :

Patients with glaucoma have a reduced quality of life in early stage disease, which worsens as the disease progresses. As I mentioned before, the proposed AI algorithm will be mainly used to identify glaucomatous optic neuropathy for screening purpose. Screening positive individuals will then be advised to see glaucoma specialists for confirmation of disease and disease management. It thus can further help with preventing disease progression, preserving vision, and improving patients’ long-term quality of life.


Q : Is there a specific message you want to deliver on the utilization of artificial intelligence in the detection of glaucomatous optic neuropathy?

Dr Carol Cheung :

Early detection and timely treatment for glaucoma can avoid vision loss. We sincerely hope that our AI work will bring significant societal impact by improving glaucoma screening in the communities.


Full article can be found at here.


You may republish this article online or in print under our Creative Commons license. You may not edit or shorten the text, you must attribute the article to 中大眼科 CUHK DOVS and you must include the author’s name in your republication.

If you have any questions, please email shaminotse@cuhk.edu.hk


Creative Commons License AttributionCreative Commons Attribution
Dr Carol Cheung was interviewed by Omnihealth Practice for her artificial intelligence screening system for glaucoma