9 Choroidal OCT Analytics
215
97 high-resolution B scans was performed with each eye, centered on the fovea. An
internal fixation light was used to center the scanning area on the fovea. Each scan
was 9.0 mm in length and spaced 30 µm apart from each other. Single OCT images
consisting of 512 A lines were acquired in 0.78 ms. The scans were obtained for
analysis after 25 frames, and averaged using built-in automatic averaging software
(TruTrack; Heidelberg Engineering, Heidelberg, Germany) to obtain a high quality choroidal image. In this work, experimental evaluation is performed on B-scans
taken from three healthy adult subjects, from whom one eye randomly chosen per
subject and 97 B-scans are taken per eye. The first two datasets has image resolution 351 × 770 and the third dataset has 496 × 1536 (covering larger area). Manual
segmentation is performed twice by same expert on each scan to study the observer
repeatability. In particular, ImageJ software is used to perform manual segmentation
[28]. The average of two such manual segmentations is taken as the reference.
9.2.2.2 Methodology
As depicted in Fig. 9.3, the SSIM-based methodology consists of various steps:
(i) denoising, (ii) localization of choriod and (iii) choroid outer boundary (COB)
detection.
9.2.2.2.1 Denoising: Generally, OCT images are noisy (Fig. 9.4a), and appropriate
denoising improves algorithmic accuracy. Accordingly, for denoising, the blockmatching and 3D filtering (BM3D) algorithm, which is generally accepted as the
state of the art, is adopted [29]. This algorithm is based on an enhanced sparse
representation in transform-domain, where enhancement of the sparsity is achieved
by grouping similar 2D image blocks into 3D data arrays called “groups”, followed
by performing collaborative filtering on them. As groups exhibit high correlation, a
decorrelating transform attenuates noise. Finally, denoised images are obtained by
applying the inverse transform (Fig. 9.4b).
9.2.2.2.2 Localization of Choroid: Next RPE inner boundary is located. This further
helps us locate the RPE outer boundary, which defines the CIB, and specify a region
of interest (ROI) between CIB and sclera which is expected to contain the COB.
Fig. 9.3 Schematic of SSIM-based methodology
215
97 high-resolution B scans was performed with each eye, centered on the fovea. An
internal fixation light was used to center the scanning area on the fovea. Each scan
was 9.0 mm in length and spaced 30 µm apart from each other. Single OCT images
consisting of 512 A lines were acquired in 0.78 ms. The scans were obtained for
analysis after 25 frames, and averaged using built-in automatic averaging software
(TruTrack; Heidelberg Engineering, Heidelberg, Germany) to obtain a high quality choroidal image. In this work, experimental evaluation is performed on B-scans
taken from three healthy adult subjects, from whom one eye randomly chosen per
subject and 97 B-scans are taken per eye. The first two datasets has image resolution 351 × 770 and the third dataset has 496 × 1536 (covering larger area). Manual
segmentation is performed twice by same expert on each scan to study the observer
repeatability. In particular, ImageJ software is used to perform manual segmentation
[28]. The average of two such manual segmentations is taken as the reference.
9.2.2.2 Methodology
As depicted in Fig. 9.3, the SSIM-based methodology consists of various steps:
(i) denoising, (ii) localization of choriod and (iii) choroid outer boundary (COB)
detection.
9.2.2.2.1 Denoising: Generally, OCT images are noisy (Fig. 9.4a), and appropriate
denoising improves algorithmic accuracy. Accordingly, for denoising, the blockmatching and 3D filtering (BM3D) algorithm, which is generally accepted as the
state of the art, is adopted [29]. This algorithm is based on an enhanced sparse
representation in transform-domain, where enhancement of the sparsity is achieved
by grouping similar 2D image blocks into 3D data arrays called “groups”, followed
by performing collaborative filtering on them. As groups exhibit high correlation, a
decorrelating transform attenuates noise. Finally, denoised images are obtained by
applying the inverse transform (Fig. 9.4b).
9.2.2.2.2 Localization of Choroid: Next RPE inner boundary is located. This further
helps us locate the RPE outer boundary, which defines the CIB, and specify a region
of interest (ROI) between CIB and sclera which is expected to contain the COB.
Fig. 9.3 Schematic of SSIM-based methodology