Abstract
Purpose :
Anterior segment (AS) OCT (optical coherence tomography) is considered a useful imaging technique to examine the AS of the eye. We propose a segmentation method based on a novel deep learning (DL) architecture for the swept source (SS) OCT volumes. The segmentation of AS images is required for quantifying these images using different techniques such as beam geometry correction and AS-OCT Angiography (AS-OCTA).
Methods :
A prototype algorithm was developed to segment the anterior and posterior corneal and iris surfaces of iridocorneal scans. We used the prototype segmentation to generate the ground truth for input images, followed by manual grading of the images to exclude segmentation errors (Fig 1). A total of 14,309 images from 53 volumes (11 subjects) with PLEX® Elite 9000 (ZEISS, Dublin, CA) with 6x6 mm cube scans (3072 pixels x 500 A-lines x 500 B-scans) were used to train a model (Bagherinia et al., IOVS June 2022, Vol.63, 2060). Augmentations, including geometrical and photometric, were applied to increase the training set to 143,090. The performance of the algorithm was evaluated using 579 images randomly selected across the 6x6 mm cube data from 29 scans (6 subjects) and reported by human manual grading. The segmentation was reviewed by human graders who rated the level of segmentation quality as being (3) acceptable, (2) partially acceptable, and (1) failed. The success rate of each segmented surface was reported.
Results :
Figure 1 shows examples of 6 mm x 6 mm OCT B-scan images and the corresponding ground truth, and examples of segmentation results overlaid with OCT images for three grading categories. Figure 2 shows a table of the success rate (quality = 3) for each segmented surface. The success rate of the anterior iris segmentation is high at 99% due to the sharp contrast at this surface. The success rate for all surfaces in the same image is 86% for grade 3, and 94% for grades to be at least 2.
Conclusions :
We prototyped a deep learning solution for AS iridocorneal segmentation using a limited number of subjects. The preliminary performance of the algorithm shows that anterior segment segmentation is valuable for generating more accurate AS-OCTA vasculature maps and may be used for beam geometry correction.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.