Page 25 - Computational Retinal Image Analysis

14 CHAPTER 2 Clinical motivation

vasodilation were 2 times more likely to have diabetic retinopathy [45].These findings
further support the concept that dynamic changes or functional changes could capture
earlier alteration even before morphological change is visible in the retinal vasculature.
There have been a variety of methods challenged to visualize and quantify retinal
blood flow. However, it has been a challenging task to be incorporated in clinical
practice. Attempts include fluorescein angiography, laser Doppler flowmetry and ve-
locimetry, ultrasound, and more recently, with Doppler OCT [46]. Recent advances
in OCT, especially its capability to capture multiple images in a very short time, have
made the Doppler OCT technique the most promising modality to quantify retinal
blood flow and other parameters non-invasively. Static visualization of the retinal
and choroidal vasculature has been achieved and equipped in current OCT machines
on the market as OCT angiography. Post processing the images acquired for OCT an-
giography has a potential to produce volumetric or velocimetric parameters. Retinal
blood flow is considered to be quite important in retinal vascular diseases such as
diabetic retinopathy, retinal vein occlusion and age-related macular degeneration,
and also glaucoma. In addition to the static morphological features, assessment of
dynamic blood flow would expand the window to detect early clinical manifestations
and to monitor disease activity including treatment response.

2 Perspectives—Precise diagnosis, replacing repetitive
work, and exploring novel signs
This chapter overviewed five key areas that have been enhanced with RIA in both
ophthalmological and systemic disease diagnosis, and prediction. RIA can enrich
clinicians to understand the pathophysiology of the disease by identifying features
in retinal images. RIA provides a non-invasive tool to probe the role of the micro-
vasculature in the development of clinical eye diseases and systemic diseases. RIA
provides additional information to stratify risks of developing diseases in the future.
By visualizing subtle changes with quantification, RIA enable clinicians to use imag-
ing markers as surrogate outcomes to monitor and evaluate response to the treatment.
There is an emerging application of imaging techniques and automated classifica-
tion or processing in RIA. There is an expectation that repetitive screening grading can
be replaced with automated grading systems without compromising high accuracy. Also
deep learning classification might provide a new insight into what clinicians do not see
in the image so far. One example might be predicting gender based on retinal images by
a deep learning model [47]. Coupled with advanced imaging modalities, RIA is a key
element to boost improved healthcare for both ophthalmology and overall healthcare.

References

[1] C.R. Keeler, The ophthalmoscope in the lifetime of Hermann von Helmholtz. Arch.
Ophthalmol. 120 (2) (2002) 194–201, https://doi.org/10.1001/archopht.120.2.194.
[2] T. Jackman, J.D. Webster, On photographing the eye of the living human retina, Phila.
Photogr. (June 5) (1886).