232
K. K. Vupparaboina et al.
9.3 Fine-Grain Analysis
In this section, we discuss fine-grain analysis of choroid. In general, such analysis can be performed in multiple ways by considering various fine features of the
choroid including choroid vascularity index, variation of choroid vessel diameter,
choroid thickness/volume variation within sublayers of choroid. However, in view
of recent developments, we restrict our focus to one such parameter namely, choroidal
stromal-luminal ratio, employed as a measure of choroidal vascularity index. To this
end, we begin by presenting the clinical underpinnings of stromal-luminal analysis.
Subsequently, we present present various solution approaches with specific focus on
recent methodology.
9.3.1 Problem Setup and Solution Approaches
So far, only two disease determinants namely, choroid thickness distribution and volume are considered for choroidal disease management. Those could even predict the
responsiveness of the retina and the choroid to anti-vascular endothelial growth factor
[42]. However, those gross indicators provide limited information about the structural changes in the choroid, and ophthalmologists seek additional information for
better understanding of diseases. Specifically, quantifying choroidal vascular region
is crucial in diagnosing diseases affecting choroid vascularity [15]. Consequently,
ophthalmologists visually inspect OCT scans paying attention to the choroid layer,
and form an opinion about the relative proportion of the constituent vessel (luminal)
and interstitial (stromal) regions (Fig. 9.12). Naturally, they seek to corroborate the
qualitative assessment with quantitative evidence.
To reduce physician’s burden, it now becomes imperative to algorithmically estimate the ratio of stromal to luminal regions from such images. Here the principal
difficulty lies in the absence of ground truth. In certain other clinical studies, such
as estimation of choroid thickness distribution, manual demarcation of the inner and
outer boundaries of the choroid is taken as the reference, albeit subject to variability
in human performance [26]. In contrast, marking choroid vessels at acceptable levels
of accuracy is practically infeasible even for experts. To appreciate this, consider the
two attempts at manual vessel segmentation, depicted in Fig. 9.13, where most larger
and some medium vessels could be marked with reasonable accuracy, but vessels
with smaller diameters were largely missed. Here the said infeasibility arises because
vessels are sometimes indicated by such extremely fine features that one cannot pinpoint individual features, but can form a gross opinion about the density of such
features. Thus it is imperative to leverage technology to identify those features.
In this context, use of the software package, ImageJ, has been widely reported
for stromal-luminal analysis [43]. Yet, while this software is versatile, it has not
been developed specifically for OCT image analysis, and does not incorporate the
idiosyncracies of this imaging modality. So, researchers used ImageJ as an interactive
K. K. Vupparaboina et al.
9.3 Fine-Grain Analysis
In this section, we discuss fine-grain analysis of choroid. In general, such analysis can be performed in multiple ways by considering various fine features of the
choroid including choroid vascularity index, variation of choroid vessel diameter,
choroid thickness/volume variation within sublayers of choroid. However, in view
of recent developments, we restrict our focus to one such parameter namely, choroidal
stromal-luminal ratio, employed as a measure of choroidal vascularity index. To this
end, we begin by presenting the clinical underpinnings of stromal-luminal analysis.
Subsequently, we present present various solution approaches with specific focus on
recent methodology.
9.3.1 Problem Setup and Solution Approaches
So far, only two disease determinants namely, choroid thickness distribution and volume are considered for choroidal disease management. Those could even predict the
responsiveness of the retina and the choroid to anti-vascular endothelial growth factor
[42]. However, those gross indicators provide limited information about the structural changes in the choroid, and ophthalmologists seek additional information for
better understanding of diseases. Specifically, quantifying choroidal vascular region
is crucial in diagnosing diseases affecting choroid vascularity [15]. Consequently,
ophthalmologists visually inspect OCT scans paying attention to the choroid layer,
and form an opinion about the relative proportion of the constituent vessel (luminal)
and interstitial (stromal) regions (Fig. 9.12). Naturally, they seek to corroborate the
qualitative assessment with quantitative evidence.
To reduce physician’s burden, it now becomes imperative to algorithmically estimate the ratio of stromal to luminal regions from such images. Here the principal
difficulty lies in the absence of ground truth. In certain other clinical studies, such
as estimation of choroid thickness distribution, manual demarcation of the inner and
outer boundaries of the choroid is taken as the reference, albeit subject to variability
in human performance [26]. In contrast, marking choroid vessels at acceptable levels
of accuracy is practically infeasible even for experts. To appreciate this, consider the
two attempts at manual vessel segmentation, depicted in Fig. 9.13, where most larger
and some medium vessels could be marked with reasonable accuracy, but vessels
with smaller diameters were largely missed. Here the said infeasibility arises because
vessels are sometimes indicated by such extremely fine features that one cannot pinpoint individual features, but can form a gross opinion about the density of such
features. Thus it is imperative to leverage technology to identify those features.
In this context, use of the software package, ImageJ, has been widely reported
for stromal-luminal analysis [43]. Yet, while this software is versatile, it has not
been developed specifically for OCT image analysis, and does not incorporate the
idiosyncracies of this imaging modality. So, researchers used ImageJ as an interactive