Deep Learning for Optic Disc Segmentation and Glaucoma Diagnosis on Retinal Images

Syna Sreng,Noppadol Maneerat,Kazuhiko Hamamoto,Khin Yadanar Win

doi:10.3390/app10144916

Abstract

Glaucoma is a major global cause of blindness. As the symptoms of glaucoma appear, when the disease reaches an advanced stage, proper screening of glaucoma in the early stages is challenging. Therefore, regular glaucoma screening is essential and recommended. However, eye screening is currently subjective, time-consuming and labor-intensive and there are insufficient eye specialists available. We present an automatic two-stage glaucoma screening system to reduce the workload of ophthalmologists. The system first segmented the optic disc region using a DeepLabv3+ architecture but substituted the encoder module with multiple deep convolutional neural networks. For the classification stage, we used pretrained deep convolutional neural networks for three proposals (1) transfer learning and (2) learning the feature descriptors using support vector machine and (3) building ensemble of methods in (1) and (2). We evaluated our methods on five available datasets containing 2787 retinal images and found that the best option for optic disc segmentation is a combination of DeepLabv3+ and MobileNet. For glaucoma classification, an ensemble of methods performed better than the conventional methods for RIM-ONE, ORIGA, DRISHTI-GS1 and ACRIMA datasets with the accuracy of 97.37%, 90.00%, 86.84% and 99.53% and Area Under Curve (AUC) of 100%, 92.06%, 91.67% and 99.98%, respectively, and performed comparably with CUHKMED, the top team in REFUGE challenge, using REFUGE dataset with an accuracy of 95.59% and AUC of 95.10%.

Highlights

The World Health Organization estimated that, in 2016, 64 million people were living with glaucoma and that it will increase to 95 million by 2030 [1]
The results indicate that the top score of methods in P1 achieves better performance for REFUGE, RIM-ONE and ACRIMA datasets comparing with the top score of methods in P2. extractors; (P2), while the top score of methods in P2 is superior compared to the top score of methods in P1 for DRISTI-GS1 dataset
The results indicate that our methods, both E(P1) and E(P2), achieve better results comparing to the conventional methods for RIM-ONE, ORIGA, DRISTI-GS1 and ACRIMA datasets with the accuracy of 97.37%, 90%, 86.84% and 99.53%, and Area Under Curve (AUC) of 100%, 92.06%, 91.67% and 99.98%, respectively, and achieve the comparable results with CUHKMED, the top team in REFUGE challenging using REFUGE dataset with the accuracy of 95.59% and AUC of 95.10%

Summary

Introduction

The World Health Organization estimated that, in 2016, 64 million people were living with glaucoma and that it will increase to 95 million by 2030 [1]. Glaucoma is an eye disease, which damages the optic nerve and can lead to blindness if left untreated. It is currently the main cause of irreversible vision loss and is caused by high intraocular pressure pushing against the optic nerve in the eye [2]. Angle-closure glaucoma causes very noticeable symptoms, for example, blurred vision, severe eye pain, sudden sight loss, light halos and more. Open-angle glaucoma slowly progresses and shows no symptoms, until peripheral vision is lost it is called “the sneak thief of sight”. Regular eye examination once per year is essential and recommended for early glaucoma screening, for people, over 40 years old, as the number of patients increases sharply with age and for people with early warning signs.

Objectives

Results

Discussion

Conclusion