Editors Selection IGR 21-3

Clinical Examination Methods: AI-assisted Chamber Angle OCT Evaluation

Tin Aung

Comment by Tin Aung on:

90793 AGE challenge: Angle Closure Glaucoma Evaluation in Anterior Segment Optical Coherence Tomography, Fu H; Li F; Sun X et al., Medical Image Analysis, 2020; 66: 101798

Find related abstracts

Angle-closure glaucoma (ACG) is a major form of glaucoma worldwide. Gonioscopy is the main clinical method for evaluating the angle to diagnose angle closure, but gonioscopy is subjective and poorly reproducible, and can be uncomfortable for patients. Physicians often use adjunctive anterior segment optical coherence tomography (AS-OCT) imaging as a quick and contactless way to image the angles in order discriminate angle closure from open angles. However AS-OCT images may need interpretation and, in most instances, one cannot reliably identify the trabecular meshwork (TM) in order to diagnose appositional angle closure, defined by iris apposition to the TM. The scleral spur is thus used as a landmark in AS-OCT scans as this structure is more easily identifiable than the TM. Angle closure is then diagnosed when there is iris apposition anterior to the scleral spur.

Although many medical image analysis algorithms have been developed for glaucoma diagnosis, few studies have focused on AS-OCT imaging. To address this, the Angle closure Glaucoma Evaluation challenge (AGE) was held in conjunction with MICCAI 2019. The AGE challenge consisted of two tasks: scleral spur localization and angle closure classification. Over 200 teams registered online, and more than 1100 results were submitted for evaluation, with eight teams participating in the final onsite challenge. In this paper, the authors summarized these eight onsite challenge methods and analyzed their corresponding results for the two tasks. The top-performing approach had an Euclidean Distance of ten pixels (10 µm) in scleral spur localization, while for the task of angle closure classification, all the algorithms achieved good performance with two best obtaining an accuracy rate of 100%. These deep learning techniques have the potential to advance the field of AS-OCT image analysis and it will be possible to automatically classify AS-OCT images as having open or closed angles. In the long term, it is hoped that new algorithms will help physicians interpret AS-OCT scans and to eventually create an 'automated gonioscopy' method that can be used to assess the angle of patients in the clinic in a non-contact, fast and reproducible way.

Issue 21-3

Change Issue