Page 103 - Computational Retinal Image Analysis

96 CHAPTER 6 Retinal vascular analysis: Segmentation, tracing, and beyond

analyzing vessel structure in retinal images. Note that there also exist artery and vein
branches that serve the rest part of the eye including cornea and iris, which are not
the main theme of this chapter.
Retinal imaging has been around for over a hundred years, with the first instrument
being invented in 1851 [9] and the first practical fundus camera developed in 1921 by
Gullstrand who was later awarded Nobel Prize for this contribution. Recently, color
fundus (CF) photography and optical coherence tomography (OCT) are among the
noninvasive and popular clinical choices.
Vasculature analysis in retinal images has, therefore, a longstanding history.
Arguably the first research effort was by Matsui et al. [10], 45 years ago, where
a mathematical morphology approach is construed toward the problem of vessel
segmentation. Overall, there has been relatively sporadic research efforts in
the 1970s and 1980s. In particular, the work of Akita and Kuga [11] considers
building up a holistic approach for understanding retinal fundus images, including
segmenting retinal blood vessels and recognizing individual artery and vein trees,
and detecting vessel structural changes such as arteriovenous nicking (AVN).
Their efforts were ambitious even by current standards, since each of the tasks
is typically pursued by a separate research paper today. The review paper by
Gilchrist [12] summarizes these early research activities. Since the 1990s, the
research activities in retinal vasculature image analysis start to proliferate,
mostly owing to the change of clinical practice in wide-spread adoption of
digital retinal imaging. Over the years, it has resulted in an enormous amount
of literature on wide-spread-related topics. In this chapter, we strive to provide
an up-to-date account, focusing more on the recent progresses and challenges
in retinal vessel analysis perspective, especially about the problems of vessel
segmentation, tracing, and classification, as, for example, shown in Fig. 1 on CF
images. Regarding the related topics, such as retinal image quality assessment,
registration and stitching of overlapping retinal images, segmentation of fovea/
macular and optic disk (OD), lesion detection and segmentation, and OCT-based
segmentation and analysis, they are covered by other chapters of this book. There
are also review articles: in Ref. [13], Abramoff and coworkers provide an excellent
overview of the history up to 2010, current status, and future perspective in both
retinal imaging and image analyses. While the survey of Kirbas and Quek [14] is
from the general perspective of segmenting and tracing vessels, the review articles
of Fraz et al. [15] and Almotiri et al. [16] focus specifically on retinal vessel
segmentation. Miri et al. [17] conduct a comprehensive study of retinal vessel
classification. Meanwhile, the reviews by Faust et al. [18] and Mookiah et al. [19]
are dedicated to image-based detecting and diagnosing of diabetic retinopathy
disease, respectively. The knowledge of retinal scans and vasculature analysis
could also be helpful in robot-assisted eye surgery [20,21]. In addition to medical
applications, retinal vasculature analysis has also been used for biometrics and
authentication purposes, see, for example, Refs. [22,23].