298
Q. Chen et al.
Inc., Dublin, CA). We performed both a qualitative and quantitative evaluations on
the dataset. To perform the qualitative evaluation, we compared both the conventional SVP and our RSVP images qualitatively for the 46 OCT scans in each of the
eight patients. To improve the visualization of conventional SVP images, we also
superimposed lesion markings which we derived from applying the techniques in
[36] to the images. We also produced SVP projection fundus images using Gorczynska’s method [38] and they were compared with RSVP images obtained from the
same datasets. Some of the patients also had color fundus photographs (CFP), which
served as the gold standard for visualizing the retina in the qualitative assessment.
Both drusen and GA were visualized on CFP. To qualitatively assess drusen visualization, we manually outlined the drusen and GA lesions in the CFP, SVP and
RSVP of these patients. We are only displaying the qualitative results of four scans
from four different patients of the total set of 46 SD-OCT scans evaluated, due to the
chapter length limitations. These displayed four scans are a representative example
of the results obtained throughout the whole dataset.
To quantitatively assess drusen visualization in SVP and RSVP images, 4 scans
from three patients were reviewed by two expert OCT readers independently. Each
independent reader marked drusen in the OCT B-scans by hand as previously
described [37]. Then, each reader independently marked every image two times
in two different sessions to enable assessment of intra-reader variation.
The gold standard for our quantitative evaluation was obtained by collapsing the
white bars along the depth axis to produce an en face drusen location map (known
as a “marking image”). For each of the image scan, the two drusen marking images
made per scan by each reader were combined using their intersection to produce a
single outline per reader per image:
R R 1 ∩ R 2
(11.1)
where R 1 and R 2 are the drusen marking images of the same scan, that are made at
two different sessions by each reader. The combined reader results were produced
by the same interpolation operation between the two reader segmentations for each
drusen outline. We also outlined the drusen by hand in the corresponding SVP and
RSVP images by each reader, as shown in Fig. 11.14. Then, the outlines produced
by SVP and RSVP were then compared quantitatively to the images marked by the
readers (the gold standard). In addition, the boundaries of drusen are very blurry in
SVP and RSVP images, making the outlines not precise. Therefore, instead of pixel
by pixel classification, we used an overlap ratio of the number of visualized drusen
as the metric to quantitatively evaluate drusen visualization in each technique:
overlap_ratio
#co_drusen
#mark_drusen
(11.2)
where ‘# co_drusen’ denotes the number of drusen outlined both in the gold standard
marked image and in the SVP or RSVP images in which outlined areas intersect
(depending which technique we were evaluating), and ‘#mark_drusen’ denotes the
Q. Chen et al.
Inc., Dublin, CA). We performed both a qualitative and quantitative evaluations on
the dataset. To perform the qualitative evaluation, we compared both the conventional SVP and our RSVP images qualitatively for the 46 OCT scans in each of the
eight patients. To improve the visualization of conventional SVP images, we also
superimposed lesion markings which we derived from applying the techniques in
[36] to the images. We also produced SVP projection fundus images using Gorczynska’s method [38] and they were compared with RSVP images obtained from the
same datasets. Some of the patients also had color fundus photographs (CFP), which
served as the gold standard for visualizing the retina in the qualitative assessment.
Both drusen and GA were visualized on CFP. To qualitatively assess drusen visualization, we manually outlined the drusen and GA lesions in the CFP, SVP and
RSVP of these patients. We are only displaying the qualitative results of four scans
from four different patients of the total set of 46 SD-OCT scans evaluated, due to the
chapter length limitations. These displayed four scans are a representative example
of the results obtained throughout the whole dataset.
To quantitatively assess drusen visualization in SVP and RSVP images, 4 scans
from three patients were reviewed by two expert OCT readers independently. Each
independent reader marked drusen in the OCT B-scans by hand as previously
described [37]. Then, each reader independently marked every image two times
in two different sessions to enable assessment of intra-reader variation.
The gold standard for our quantitative evaluation was obtained by collapsing the
white bars along the depth axis to produce an en face drusen location map (known
as a “marking image”). For each of the image scan, the two drusen marking images
made per scan by each reader were combined using their intersection to produce a
single outline per reader per image:
R R 1 ∩ R 2
(11.1)
where R 1 and R 2 are the drusen marking images of the same scan, that are made at
two different sessions by each reader. The combined reader results were produced
by the same interpolation operation between the two reader segmentations for each
drusen outline. We also outlined the drusen by hand in the corresponding SVP and
RSVP images by each reader, as shown in Fig. 11.14. Then, the outlines produced
by SVP and RSVP were then compared quantitatively to the images marked by the
readers (the gold standard). In addition, the boundaries of drusen are very blurry in
SVP and RSVP images, making the outlines not precise. Therefore, instead of pixel
by pixel classification, we used an overlap ratio of the number of visualized drusen
as the metric to quantitatively evaluate drusen visualization in each technique:
overlap_ratio
#co_drusen
#mark_drusen
(11.2)
where ‘# co_drusen’ denotes the number of drusen outlined both in the gold standard
marked image and in the SVP or RSVP images in which outlined areas intersect
(depending which technique we were evaluating), and ‘#mark_drusen’ denotes the