Abstract
Purpose :
Swept source optical coherence tomography offers several advantages including faster scan rates and deeper imaging into the choroid. While these two parameters are used to expand the field of view and increase the density of the scan, the amount of data that needs to be processed increases rapidly. The purpose of this study is to determine an acceptable trade-off between segmentation processing time and the quality of segmentation for the retinal pigment epithelium (RPE).
Methods :
METHODS:
Three eyes with dry age-related macular degeneration (AMD) were imaged on PLEX™ Elite 9000 (ZEISS, Dublin, CA). For each eye, a 12mm x 12 mm cube consisting of 1024 B-scans and 1024 A-scans per B-scan was acquired and processed for segmentation as acquired and after down-sampling the acquired cube by a factor of 2 and 4 (denoted as x2 and x4); which resulted in 22 um and 44 um resolutions respectively. The quality of segmentation was compared by reviewing the RPE segmentation overlaid on the B-scans in the presence of drusen and by analyzing morphologic features present on a slab defined from the RPE to the Bruch’s membrane.
Results :
Image 1 shows the original RPE segmentation and segmentation produced by down-sampling the cube by factors of 2 and 4 and overlaid on a B-scan with evident irregularities in the RPE. The down-sampling factor of 2 method remains accurate in segmenting lesions affecting the RPE with only minor differences in some regions. The x4 down-sampling method produces gross segmentation errors that underestimates the size of the lesions. The elevated RPE profile produced from the original segmentation and from the two down-sampling methods are shown on images 2A, 2B, and 2C. If we ignore the artifacts from the difference in segmenting the optic disc and concentrate on the macular region where drusen are present, the root mean square value of the difference in RPE elevation from the original RPE elevation is 6 um and 12 um for the x2 and x4 down-sampling method respectively.
Conclusions :
By comparing the segmentations overlaid on the B-scans, we found that the x2 down-sampling method remains accurate in segmenting the RPE. This method speeds up the algorithm by a factor of 4 and represents a good trade-off between quality of the RPE segmentation and processing time.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.