Computer Vision News - February 2022

52 Medical Imaging Challenge The AIROGS Challenge is about the automated detection of an eye disease called glaucoma , which affects the optic disc in the eye and can cause blindness. Usually, patients are identified through color fundus photography graded by human experts , but this can be a slow and inefficient process with so many of these images to analyze. To solve this, AI approaches have been proposed to automate the screening process and make it more cost-efficient. These have proved promising in the lab; however, they do not perform as well in real-world scenarios . This challenge aims to develop automated methods with a focus on screening performance and robustness of results . Participants have been given a training set with gradable images for developing their models. There is also a closed test set including ungradable images to encourage the development of methods with robustness mechanisms to detect out-of- distribution samples . Methods cannot be trained with the ungradable images. “ We have a large dataset, which we want to be as close to a real-world scenario as possible, ” Coen explains. “ It includes data from many different screening sites, optometrists, and hospitals. All images have been manually graded with a label of either referable glaucoma, non-referable glaucoma, or ungradable. We wanted a truly diverse dataset that will result in much more robust solutions than already exist. ” AIROGS: ARTIFICIAL INTELLIGENCE FOR ROBUST GLAUCOMA SCREENING CHALLENGE Coen de Vente is a third-year PhD student at the University of Amsterdam, under the supervision of Full Professor of AI and Health Clarisa Sanchez. They are two of the co-organizers of AIROGS, the Artificial Intelligence for Robust Glaucoma Screening Challenge, which is taking place as we speak and will present its results at ISBI 2022 in India next month. Coen and Clarisa tell us more about the challenge as it enters its final test phase. Coen de Vente Clarisa Sanchez