Page 144 - Computational Retinal Image Analysis

138 CHAPTER 8 Image quality assessment

programs. Screening for diabetic retinopathy (DR) is the most well-known and
established screening system, but automated diagnostic systems are emerging to
address screening requirements for diseases such as glaucoma [7]. DR screening
programs are covered more fully elsewhere in the book, hence the following summary
provides an image quality assessment perspective.
One in ten adults were estimated to have diabetes in 2014 [8], and numbers are
increasing. Given sight-loss caused by diabetic retinopathy is a common consequence
of the diabetes, regular eye screening among the diabetic population is an effective way
to reduce sight loss caused by diabetic retinopathy. Screening for diabetic retinopathy
can reduce the risk of blindness by less than half compared with unscreened controls
[9]. National screening programs, such as the UK National Health Service Diabetic
Eye Screening Programme (NHS DESP) [10] generally rely upon the assessment of
images performed remotely from the point of capture. Currently in the NHS DESP,
this is performed by human graders at reading centers. Screening programs face
specific challenges related to image quality. The image capture process is tightly
controlled with defined camera types and image capture protocols. However, because
the images are often taken at different sites, the trained ophthalmic photographers
may have various degrees of experience, which can affect the quality of image
capture. Moreover, subjects may have differing degrees of lenticular/media opacities
or subjects may move while the image is being taken, which can lead to images
of inferior quality being collected from screening programs. It has been reported
that in some cases 12% of total images in a screening program are low-quality and
ungradable [11]. One of the aims of a diabetic retinopathy screening system is to
detect the first signs of diabetic retinopathy. A pathognomonic sign of early diabetes
is the presence of microaneurysms, which are small and difficult to see. If an image
is low quality these subtle signs can be missed causing the patient to be incorrectly
diagnosed as disease free, i.e., a false negative. A crucial opportunity will therefore
be missed to prescribe treatment or lifestyle changes that could slow down or halt the
progression of diabetic retinopathy.
In the NHS DESP, images are graded in terms of severity of retinopathy if the
image is of adequate quality. In general, image clarity across the entire image is
important to ensure subtle abnormalities are detected. If an image is low quality
or “inadequate” this is also recorded in a systematic manner. An image is graded
as adequate if “the area of interest is positioned within the defined limits and has
enough clarity for a reliable decision on the patient’s status” [12]. Within the English
NHS DESP, one optic disc centered image and one fovea centered image, both with
field of view 45°, are captured for each eye of the subject. The correct position of the
image is defined in terms of correct centration of the area of interest. For example,
the optic disc centered image is classified as adequate if a complete optic disc can be
seen in the image which should be more than two optic disc diameters from the edge
of the image. In addition, fine vessels should be visible on the surface of the disc. An
image is graded as inadequate if the area of interest is not correctly positioned and/or
the clarity of the image is not sufficient for a reliable decision on retinopathy grade
to be made [12].