270
F. Shi et al.
10.4.2.5 Post-processing
The vessel silhouettes in the EZ have lower values of intensity, and the voxels in
these regions may be falsely classified as disrupted. The vessel silhouettes are identified and detected based on a vessel detector [52]. As in the outer retina (EZ to
RPE), the vessel silhouettes offer excellent contrast; only those voxels between the
EZ and RPE are selected and each pixel in the 2D projection image is the average
in the z-axis direction of the selected voxels at that particular x, y location in the
OCT volume. Then, the vessel silhouettes are segmented using a KNN classifier.
If the detected EZ disruption regions have the same x and y location as the vessel
silhouettes, these regions are regarded as normal and removed as false detections.
Due to the physiological connectivity of the EZ disrupted/non-disrupted regions, isolated disrupted/non-disrupted voxels are eliminated through morphological opening
operations, where the shape of the structural element is set as ball with a radius of 5
voxels.
10.4.3 Results
In total, 15 eyes in subjects with retinal trauma and 15 eyes in normal subjects
were included and underwent a macular-centred (6 × 6 mm) SD-OCT scan (Topcon
3D OCT-1000, 512 × 64 × 480 voxels, 11.72 × 93.75 × 3.50 µm
3 , or 512 × 128 ×
480 voxels, 11.72 × 46.88 × 3.50 µm
3 ). There were 12 males and 3 females in the
trauma group, with a mean age of 30.3 ± 11.3 years (range: 8–43 years). There were
9 males and 6 females in the normal group, with a mean age of 33.1 ± 10.8 years
(range: 7–46 years). Subjects with other eye diseases were excluded except for those
with refractive error <= ±6 diopter.
The Institutional Review Board of the Joint Shantou International Eye Center
approved this study and waived informed consent due to the retrospective nature of
this study. Our study also complies with the Declaration of Helsinki. The patient
records/information was made anonymous prior to analysis.
To evaluate the performance of the proposed method, all the EZ disruption regions
in the 3D SD-OCT images were manually marked by an ophthalmologist slice by slice
using the ITK-SNAP software [75] and saved as the ground truth. The leave-one-out
method was used to train the Adaboost based integrated classifier models. Because the
sample ratio of the majority class (non-disrupted) and the minority class (disrupted)
was approximately (110 ± 256):1 on average, non-disrupted samples were randomly
selected to match the disrupted ones. The EZ disruption volume was calculated by
multiplying the disruption number by the voxel resolution.
The mean and 95% confidence intervals of the segmented EZ disruption region
volumes were compared between eyes with retinal trauma and normal eyes. Student’s t-test was used to evaluate the statistical significance of the disruption volume differences between the two groups of eyes. Statistical correlation analysis and
F. Shi et al.
10.4.2.5 Post-processing
The vessel silhouettes in the EZ have lower values of intensity, and the voxels in
these regions may be falsely classified as disrupted. The vessel silhouettes are identified and detected based on a vessel detector [52]. As in the outer retina (EZ to
RPE), the vessel silhouettes offer excellent contrast; only those voxels between the
EZ and RPE are selected and each pixel in the 2D projection image is the average
in the z-axis direction of the selected voxels at that particular x, y location in the
OCT volume. Then, the vessel silhouettes are segmented using a KNN classifier.
If the detected EZ disruption regions have the same x and y location as the vessel
silhouettes, these regions are regarded as normal and removed as false detections.
Due to the physiological connectivity of the EZ disrupted/non-disrupted regions, isolated disrupted/non-disrupted voxels are eliminated through morphological opening
operations, where the shape of the structural element is set as ball with a radius of 5
voxels.
10.4.3 Results
In total, 15 eyes in subjects with retinal trauma and 15 eyes in normal subjects
were included and underwent a macular-centred (6 × 6 mm) SD-OCT scan (Topcon
3D OCT-1000, 512 × 64 × 480 voxels, 11.72 × 93.75 × 3.50 µm
3 , or 512 × 128 ×
480 voxels, 11.72 × 46.88 × 3.50 µm
3 ). There were 12 males and 3 females in the
trauma group, with a mean age of 30.3 ± 11.3 years (range: 8–43 years). There were
9 males and 6 females in the normal group, with a mean age of 33.1 ± 10.8 years
(range: 7–46 years). Subjects with other eye diseases were excluded except for those
with refractive error <= ±6 diopter.
The Institutional Review Board of the Joint Shantou International Eye Center
approved this study and waived informed consent due to the retrospective nature of
this study. Our study also complies with the Declaration of Helsinki. The patient
records/information was made anonymous prior to analysis.
To evaluate the performance of the proposed method, all the EZ disruption regions
in the 3D SD-OCT images were manually marked by an ophthalmologist slice by slice
using the ITK-SNAP software [75] and saved as the ground truth. The leave-one-out
method was used to train the Adaboost based integrated classifier models. Because the
sample ratio of the majority class (non-disrupted) and the minority class (disrupted)
was approximately (110 ± 256):1 on average, non-disrupted samples were randomly
selected to match the disrupted ones. The EZ disruption volume was calculated by
multiplying the disruption number by the voxel resolution.
The mean and 95% confidence intervals of the segmented EZ disruption region
volumes were compared between eyes with retinal trauma and normal eyes. Student’s t-test was used to evaluate the statistical significance of the disruption volume differences between the two groups of eyes. Statistical correlation analysis and