Primary open-angle glaucoma diagnosis from fundusphotographs using siamese network

MINGQUAN LIN, Lei Liu, Mae Gordon, Michael Kass, Fei Wang, Sarah H. Van Tassel, Yifan Peng

Keywords: Deep learning, Primary open-angle glaucoma (POAG), Fundus photographs, POAG onsite detection

TL;DR: We present an automated computer-based classification algorithm using variable fundus photographs.

Abstract: Primary open-angle glaucoma (POAG) is one of the leading causes of irreversible blindness in the United States and worldwide. Although deep learning methods have been proposed to diagnose POAG, these methods all used a single image as input. Differently, the glaucoma specialists compare the follow-up image with the baseline image to determine a glaucomatous eye. To simulate this process, we proposed a siamese network model, POAGNet, to identify POAG from fundus photographs. The POAGNet consists of two side-outputs for deep supervision. The POAGNet network was trained and evaluated on two datasets: (1) 37,339 fundus photographs from 1,636 Ocular Hypertension Treatment Study (OHTS) participants, and (2) 3,684 fundus photographs from the sequential fundus images for glaucoma (SIG) dataset. Extensive experiments show that POAGNet performed better on POAG diagnosis in the OHTS test set with an accuracy of 0.91, F-score of 0.5069, and an AUC of 0.9081 than state-of-the-art (accuracy 0.8320; F-score 0.3864; AUC 0.8750). It also outperformed the baseline in the SIG dataset (Accuracy 0.9176 vs 0.8690; F-score 0.1613 vs 0.1010; AUC 0.7518 vs 0.6434). These results highlight the potential of deep learning to assist and enhance clinical POAG diagnosis. The proposed network will be publicly available on \url{https://github.com/bionlplab/poagnet}.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.

Paper Type: both

Primary Subject Area: Detection and Diagnosis

Secondary Subject Area: Detection and Diagnosis

Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.