11 Segmentation and Visualization of Drusen …
301
Fig. 11.13 Comparison of OCT retinal image projections. SVP fundus image (a). RSVP image (b).
Projection fundus images (c, d, e, f) using Gorczynska’s method [38] which are the outer nuclear
layer level, photoreceptor outer segment level, RPE level, and choroid level, respectively. g, h, and
i are three B-scans corresponding to the red, blue and yellow lines in the projection images (a–f).
The red triangle shows a blood vessel in corresponding images and the green and blue triangles
show drusen with high and low reflectivity, respectively
It can be seen that both drusen can be more clearly identified in the RSVP image
than in any other individual image.
Figure 11.14 shows drusen boundaries manually outlined by referring to B-scans.
Figure 11.14g-i show three B-scans corresponding to the red, blue and yellow lines
in Fig. 11.37d–f, respectively. Drusen and GA were easily differentiated in B-scans.
The blue and green outlines represent the larger drusen and GA, respectively. Red,
blue and yellow triangles delineate corresponding locations of GA in the CFP (14D),
SVP (14E), RSVP (14F), and B-scan images (14G-I). GA extent can be identified in
the SVP image (14E). However, drusen appear more difficult to visualize with SVP.
From Fig. 11.14, it can be observed that nearly all drusen that were outlined on the
CFP were also observed in the RSVP image.
For reader 1 and 2, the mean intra-reader overlap ratio measured between two
segmentation sessions of the B-scans were 67.04 and 90.88%. While the mean interreader overlap ratio measured between the B-scan segmentations of two different
readers was 65.14%. Table 11.1 shows the drusen overlap ratio for a method (SVP or
RSVP) and the gold standard (B-scan segmentation) in four OCT scans from three
different patients. The overlap ratio is evidently higher for RSVP than with SVP.
Most of the drusen were not effectively visualized with SVP images; as such, they
were completely absent in scans 3 and 4. Most of the drusen were clearly visible in
the RSVP images. The “Readers 1 and 2” column corresponds to the evaluation of
the combined segmentations of both readers.
11.2.2.6 Discussion
The ability of RSVP images and their quality to visualize retinal drusen were superior
to those known conventional SVP images. More so, GA is not visualized with the
RSVP method because GA occurs beneath the RPE layers and is not included in the
projection images, thereby eliminating a confounding factor for visualizing drusen
that is present in conventional SVP images. The RSVP method also gives more
accurate characterization of drusen thickness than the color marking method [37]
that only defines the location of drusen. The conventional SVP was not able to
adequately visualize drusen, but the Gorczynska’s method [38] was able to display
drusen in fundus images. Hence, drusen were divided up into several fundus images
at different anatomic levels. The RSVP method enhanced the visualization of drusen
by utilizing a comprehensive single image. In addition, blood vessel visualization
was also more effective with the RSVP method because the projected region was
301
Fig. 11.13 Comparison of OCT retinal image projections. SVP fundus image (a). RSVP image (b).
Projection fundus images (c, d, e, f) using Gorczynska’s method [38] which are the outer nuclear
layer level, photoreceptor outer segment level, RPE level, and choroid level, respectively. g, h, and
i are three B-scans corresponding to the red, blue and yellow lines in the projection images (a–f).
The red triangle shows a blood vessel in corresponding images and the green and blue triangles
show drusen with high and low reflectivity, respectively
It can be seen that both drusen can be more clearly identified in the RSVP image
than in any other individual image.
Figure 11.14 shows drusen boundaries manually outlined by referring to B-scans.
Figure 11.14g-i show three B-scans corresponding to the red, blue and yellow lines
in Fig. 11.37d–f, respectively. Drusen and GA were easily differentiated in B-scans.
The blue and green outlines represent the larger drusen and GA, respectively. Red,
blue and yellow triangles delineate corresponding locations of GA in the CFP (14D),
SVP (14E), RSVP (14F), and B-scan images (14G-I). GA extent can be identified in
the SVP image (14E). However, drusen appear more difficult to visualize with SVP.
From Fig. 11.14, it can be observed that nearly all drusen that were outlined on the
CFP were also observed in the RSVP image.
For reader 1 and 2, the mean intra-reader overlap ratio measured between two
segmentation sessions of the B-scans were 67.04 and 90.88%. While the mean interreader overlap ratio measured between the B-scan segmentations of two different
readers was 65.14%. Table 11.1 shows the drusen overlap ratio for a method (SVP or
RSVP) and the gold standard (B-scan segmentation) in four OCT scans from three
different patients. The overlap ratio is evidently higher for RSVP than with SVP.
Most of the drusen were not effectively visualized with SVP images; as such, they
were completely absent in scans 3 and 4. Most of the drusen were clearly visible in
the RSVP images. The “Readers 1 and 2” column corresponds to the evaluation of
the combined segmentations of both readers.
11.2.2.6 Discussion
The ability of RSVP images and their quality to visualize retinal drusen were superior
to those known conventional SVP images. More so, GA is not visualized with the
RSVP method because GA occurs beneath the RPE layers and is not included in the
projection images, thereby eliminating a confounding factor for visualizing drusen
that is present in conventional SVP images. The RSVP method also gives more
accurate characterization of drusen thickness than the color marking method [37]
that only defines the location of drusen. The conventional SVP was not able to
adequately visualize drusen, but the Gorczynska’s method [38] was able to display
drusen in fundus images. Hence, drusen were divided up into several fundus images
at different anatomic levels. The RSVP method enhanced the visualization of drusen
by utilizing a comprehensive single image. In addition, blood vessel visualization
was also more effective with the RSVP method because the projected region was