9 Choroidal OCT Analytics
227
Specifically, two quotients are defined. The quotient of mean, QMz, is defined
by the ratio
QMz =
|Mz
auto
− z
ideal
|
|Mz ref − z ideal |
,
(9.7)
where Mz
auto and Mz
ref , respectively, indicate the mean values obtained by the
algorithm and the manual method, and z
ideal denotes the ideal value of z. A
low QMz value is desirable. Specifically, QMz = 1 would make the algorithmic
accuracy indistinguishable from the accuracy of manual methods in terms of
mean error. Similarly, quotient of CV, QCVz, is defined by
QCVz =
CVz
auto
CVz ref ,
(9.8)
where CVz
auto and CVz
ref , respectively, indicate the CV obtained by the algorithm and that obtained manually. Again, we desire QCVz, measuring relative
standard dispersion, to be low, and QCVz = 1 would make the algorithm at par
with manual methods. In (9.7) and (9.8), the general quantity z can specifically
be either AD, or CC, or DC, as mentioned earlier.
In this backdrop, the respective overall QMAD and overall QCVAD values,
obtained by SSIM-based method, are observed to be, 1.59 and 0.93. Interestingly, QCVAD value for SSIM-based algorithm is less than one, indicating that
the algorithmic consistency exceeds manual consistency. Further, those quotient values improve upon the respective quotients of 2.03 and 1.69 reported
by Alonso-Caneiro et al. [19] by respective factors of 27.67% and 81.72%.
Figure 9.7 provides further dataset-wise details. In particular, the SSIM-based
method achieves consistent QMAD values of 1.55, 1.87, 1.35 for the three
datasets, indicating consistent algorithmic performance across datasets. Such
performance consistency is further buttressed by the corresponding consistent
QCVAD values of 0.84, 0.83 and 1.22. In addition, the low overall QMCC and
QCVCC (resp. QMDC and QCVDC) values of 2 and 1.94 (resp. 1.63 and 1.37),
respectively, further corroborate the effectiveness of the SSIM-based algorithm.
Figure 9.7 presents further details on dataset-wise performance.
9.2.3.2 Choroidal Volume
Finally, estimation choroidal volume is attempted. As mentioned in Sect. 9.2.2.1, each
of the three datasets consists of 97 B-scans taken at a uniform vertical separation
of 30 µm. As these datasets were taken from in vivo imaging, all the scans are not
spatially aligned, specifically along Z-axis, due to the eye movement. Therefore,
choroid layer is not aligned spatially in all the scans and thus volume could not be
estimated directly. In view of this, first the scans are geometrically aligned, thereby
aligning the choroid layer. In particular, building on eye structure, spherical alignment
227
Specifically, two quotients are defined. The quotient of mean, QMz, is defined
by the ratio
QMz =
|Mz
auto
− z
ideal
|
|Mz ref − z ideal |
,
(9.7)
where Mz
auto and Mz
ref , respectively, indicate the mean values obtained by the
algorithm and the manual method, and z
ideal denotes the ideal value of z. A
low QMz value is desirable. Specifically, QMz = 1 would make the algorithmic
accuracy indistinguishable from the accuracy of manual methods in terms of
mean error. Similarly, quotient of CV, QCVz, is defined by
QCVz =
CVz
auto
CVz ref ,
(9.8)
where CVz
auto and CVz
ref , respectively, indicate the CV obtained by the algorithm and that obtained manually. Again, we desire QCVz, measuring relative
standard dispersion, to be low, and QCVz = 1 would make the algorithm at par
with manual methods. In (9.7) and (9.8), the general quantity z can specifically
be either AD, or CC, or DC, as mentioned earlier.
In this backdrop, the respective overall QMAD and overall QCVAD values,
obtained by SSIM-based method, are observed to be, 1.59 and 0.93. Interestingly, QCVAD value for SSIM-based algorithm is less than one, indicating that
the algorithmic consistency exceeds manual consistency. Further, those quotient values improve upon the respective quotients of 2.03 and 1.69 reported
by Alonso-Caneiro et al. [19] by respective factors of 27.67% and 81.72%.
Figure 9.7 provides further dataset-wise details. In particular, the SSIM-based
method achieves consistent QMAD values of 1.55, 1.87, 1.35 for the three
datasets, indicating consistent algorithmic performance across datasets. Such
performance consistency is further buttressed by the corresponding consistent
QCVAD values of 0.84, 0.83 and 1.22. In addition, the low overall QMCC and
QCVCC (resp. QMDC and QCVDC) values of 2 and 1.94 (resp. 1.63 and 1.37),
respectively, further corroborate the effectiveness of the SSIM-based algorithm.
Figure 9.7 presents further details on dataset-wise performance.
9.2.3.2 Choroidal Volume
Finally, estimation choroidal volume is attempted. As mentioned in Sect. 9.2.2.1, each
of the three datasets consists of 97 B-scans taken at a uniform vertical separation
of 30 µm. As these datasets were taken from in vivo imaging, all the scans are not
spatially aligned, specifically along Z-axis, due to the eye movement. Therefore,
choroid layer is not aligned spatially in all the scans and thus volume could not be
estimated directly. In view of this, first the scans are geometrically aligned, thereby
aligning the choroid layer. In particular, building on eye structure, spherical alignment