1
Disc-aware Ensemble Network for Glaucoma
Screening from Fundus Image
Huazhu Fu, Jun Cheng, Yanwu Xu, Changqing Zhang, Damon Wing Kee Wong, Jiang Liu, and Xiaochun Cao
Abstract—Glaucoma is a chronic eye disease that leads to
irreversible vision loss. Most of the existing automatic screening
methods firstly segment the main structure, and subsequently
calculate the clinical measurement for detection and screening
of glaucoma. However, these measurement-based methods rely
heavily on the segmentation accuracy, and ignore various visual
features. In this paper, we introduce a deep learning technique to
gain additional image-relevant information, and screen glaucoma
from the fundus image directly. Specifically, a novel Disc-aware
Ensemble Network (DENet) for automatic glaucoma screening
is proposed, which integrates the deep hierarchical context of
the global fundus image and the local optic disc region. Four
deep streams on different levels and modules are respectively
considered as global image stream, segmentation-guided network,
local disc region stream, and disc polar transformation stream.
Finally, the output probabilities of different streams are fused
as the final screening result. The experiments on two glaucoma
datasets (SCES and new SINDI datasets) show our method
outperforms other state-of-the-art algorithms.
Index Terms—Deep learning, glaucoma screening, optic disc
segmentation, neural network.
I. INTRODUCTION
Glaucoma is one of the major leading causes of blindness
among eye diseases, predicted to affect around 80 million
people by 2020 [1]. Unlike other eye diseases such as cataracts
and myopia, vision loss from glaucoma cannot be reversed.
Early screening is thus essential for early treatment to preserve
vision and maintain life quality. However, many glaucoma
patients are not aware of their condition [2]. That is why
glaucoma is also called the “silent theft of sight”, as shown
in the bottom row of Fig. 1. Clinically, there are three ex-
aminations practiced to screen glaucoma: intraocular pressure
(IOP) measurement, function-based visual field test, and optic
nerve head (ONH) assessment. IOP is an important risk factor
but not specific enough to be an effective detection tool for
a great number of glaucoma patients with normal tension.
Function-based visual field testing requires specialized peri-
metric equipment not normally present in primary healthcare
H. Fu and D. W. K. Wong are with Institute for Infocomm Research,
Agency for Science, Technology and Research, Singapore 138632 (e-mail:
huazhufu@gmail.com, wkwong@i2r.a-star.edu.sg).
J. Cheng and J. Liu are with the Cixi Institute of Biomedical Engi-
neering, Chinese Academy of Sciences, Zhejiang 315201, China (e-mail:
sam.j.cheng@gmail.com, jimmyliu@nimte.ac.cn)
Y. Xu is with CVTE Research, Guangzhou Shiyuan Electronic Technology
Company Limited, Guangzhou 510530, China (e-mail: xuyanwu@cvte.com).
C. Zhang is with the School of Computer Science and Technology, Tianjin
University, Tianjin 300072, China (e-mail: zhangchangqing@tju.edu.cn)
X. Cao is with the State Key Laboratory of Information Security, Institute
of Information Engineering, Chinese Academy of Sciences, Beijing 100093,
China (e-mail: caoxiaochun@iie.ac.cn)
VDD
VCD
Fundus Image Normal Glaucoma
Neuroretinal
Rim
No Vision Loss (Normal) Vision Loss (Glaucoma)
Fig. 1. Top: the whole fundus image and zoom-in normal/glaucoma disc
regions, where the vertical cup to disc ratio (CDR) is calculated by the ratio
of vertical cup diameter (VCD) to vertical disc diameter (VDD). Bottom: the
visual fields with normal and glaucoma cases.
clinics. Moreover, the early glaucoma often does not have
visual symptoms. ONH assessment is a convenient way to
detect glaucoma early, and is currently performed widely by
trained glaucoma specialists [3]–[5].
Manual ONH assessment by trained clinicians is time-
consuming and costly. Thus, an automatic method is necessity
for screening. One popular ONH assessment method is based
on the measurement of clinical parameters, such as the vertical
cup to disc ratio (CDR) [6], rim to disc area ratio, and
disc diameter [7]. Among them, CDR is well accepted and
commonly used by clinicians. As shown in the top row of
Fig. 1, the CDR is calculated by the ratio of vertical cup
diameter (VCD) to vertical disc diameter (VDD). In general, a
larger CDR suggests a higher risk of glaucoma and vice versa.
For automatic screening, measurement-based methods have
been proposed [8]–[12], which segment the main structure
(e.g., optic disc and optic cup) first, and then calculate the
clinical measurement values to identify the glaucoma cases.
For example, a superpixel-based classifier with various hand-
crafted visual features is utilized to extract the optic disc
and cup regions [10]. The CDR value is then calculated
based on the segmented regions. In [13], a CDR assessment
using fundus image is proposed, where a sparse dissimilarity-
constrained coding approach is employed to consider both
the dissimilarity constraint and the sparsity constraint from
a set of reference discs with known CDRs. The reconstruction
arXiv:1805.07549v1 [cs.CV] 19 May 2018