Page 145 - Computational Retinal Image Analysis

1 Introduction 139

Because screening programs have to collect and store large numbers of images,
compression techniques are selected that reduce the image size from over 20 MB down
to 1–2 MB without loss of image quality and any clinically significant information
[13]. The uncompressed images taken in the screening programs often rely on
pharmacologically dilated pupils (mydriasis) to maximize image quality. The impact
of mydriasis on accuracy in assessment of images was reported by Hansen et al.
[14]. The use of mydriasis in screening programs is not universal. Routine mydriasis
is implemented for UK NHS DESP, however screening programs elsewhere may
operate without routine mydriasis.
Algorithms that are aimed at automating the diabetic retinopathy grading process
are emerging [8, 15–18], which will enable grading to be performed with automated
software. Image quality assessments are increasingly being used as part of this
automation process. With respect to mydriasis, this is often undertaken in diabetic
screening systems and can affect image quality. However, Gulshan et al. [16], reports
that the performance of their algorithm does not drop significantly when analyzing
images captured with and without pharmacological pupil dilation. Further to the
diabetic screening programs, progress is also being made to produce automated
methods to assist with other disease-screening programs such as glaucoma, macular
degeneration and retinopathy of prematurity [7]. Retinopathy of prematurity is
a condition that affects pre-term infants and algorithms have focused on retinal
vasculometry assessment [19].
1.2.2 Teleophthalmology and clinical decision making
A further example of where image quality relating to diagnostic criteria requires
consideration relates to teleophthalmology where the use of portable imaging
systems can often lead to wide variations in image quality during routine clinical
use. With the arrival of smaller and more portable retinal imaging systems, ranging
from portable cameras (e.g. EpiCam M by Epipole Ltd.) to lens attachments for use
with mobile phones (e.g. Peek Retina by Peek Vision Ltd.), telemedicine in the form
of teleophthalmology can be realized by sending ophthalmic images to specialist
centers, to obtain expert opinion on a range of conditions. This technique offers an
efficient and cost-effective use of medical resource in communities that may not
have any access to expert opinion [20]. Teleophthalmology can also contribute to
screening programs such as diabetic retinopathy [21] by expanding the availability
of the service from existing established screening centers, to populations with less
access to health services.
In addition to diabetic retinopathy, teleophthalmology systems have been
explored in relation to retinopathy of prematurity (ROP) [22] where images acquired
by a neonatal nurse have been compared to those acquired by an experienced
ophthalmologist. Giancardo et al. [23] describes a telemedicine network that
performed teleophthalmology in the US, which provided screening for diabetic
retinopathy and other diseases of the retina.
Image quality is a vital aspect to be considered in any teleophthalmology system
due to the remoteness of image capture and the asynchronous nature of image capture