A Deep Learning System for Classification of Papilledema Severity on Ocular Fundus Photographs

Papilledema severity is an important predictive factor for visual outcomes of patients with intracranial hypertension. The aim of our study was to evaluate the performance of an artificial intelligence deep learning system (DLS) to classify the severity of papilledema on standard mydriatic retinal fundus photographs.

Poster 106 A Deep Learning System For Classification Of Papilledema Severity On Ocular Fundus Photographs Caroline Vasseneix1, Raymond Najjar1, Xinxing Xu2, Zhiqun Tan1, Jing Liang Loo3, Shweta Singhal3, Sharon Tow3, Leonard Milea4, Daniel Ting3, Liu Yong2, Tien Yin Wong3, Nancy Newman5, Valerie Biousse5, Dan Milea1 Singapore Eye Research Institute, Singapore, Singapore, 2Institute of High Performance Computing (A*STAR), Singapore, Singapore, Singapore National Eye Center, Singapore, Singapore, 4UC Berkeley, Berkeley, California, USA, 5Emory School of Medicine, Atlanta, Georgia, USA 1 3 Introduction: Papilledema severity is an important predictive factor for visual outcomes of patients with intracranial hypertension. The aim of our study was to evaluate the performance of an artificial intelligence deep learning system (DLS) to classify the severity of papilledema on standard mydriatic retinal fundus photographs. Methods: A DLS was trained to automatically classify papilledema severity in 965 patients with confirmed intracranial hypertension, in a multiethnic cohort (BONSAI – Brain and Optic Nerve Study with Artificial Intelligence). Training was performed on 1052 photographs of mild/moderate papilledema (MP, Frisén grades 1-3, no major vessel obscuration on the optic disc) and 1051 photographs of severe papilledema (SP, Frisén grades 4-5, any vessel obscuration on the optic disc), classified by a panel of experts on the basis of a modified Frisén scale. Subsequently, the performance of the DLS and that of three independent neuro-ophthalmologists was tested in 111 patients (214 photographs, 92 with MP and 122 with SP), by calculating the area under the receiver operating characteristics curve (AUC), accuracy, sensitivity and specificity. Kappa agreement scores between 1/ the DLS and each of the three neuroophthalmologist and 2/ the three neuro-ophthalmologists were calculated. Results: The DLS could successfully discriminate between photographs of MP and SP, with an AUC of 0.93 (95%CI: 0.89-0.96) and an accuracy, sensitivity and specificity of 87.9%, 91.8% and 86.2%, respectively. This performance was comparable with the majority agreement (when at least 2 graders agreed on the classification) between the three neuro-ophthalmologists (84.1%, 91.8% and 73.9%, P=0.19, P=1, P=0.09, respectively). Agreement scores between the DLS and the neuro-ophthalmologists’ evaluation was 0.62 (CI 95% 0.57-0.68), and inter-grader agreement between the three graders was 0.54 (CI 95% 0.47-0.62). Conclusions: Our DLS accurately classified the severity of papilledema on an independent set of mydriatic fundus photographs, achieving a comparable performance with that of 3 neuro-ophthalmologists. References: None. Keywords: high intracranial pressure/headache, pseudotumor cerebri Financial Disclosures: The authors had no disclosures. Grant Support: Supported by the Singapore National Medical Research Council (Clinician Scientist Individual Research grant CIRG18Nov-0013), and the Duke-NUS Medical School, Ophthalmology and Visual Sciences Academic Clinical Program grant (05/FY2019/P2/06-A60). Contact Information: Caroline Vasseneix, [e-mail] 264 | North American Neuro-Ophthalmology Society