Purpose:
Spectral Domain Optical Coherence Tomography enables depth-resolved imaging of the retina which is critical for the quantification of retinal layer thicknesses. Manual segmentation of these layers is labor intensive. A fully automated method to segment and quantify retinal layers would reduce labor and time costs and provide an objective evaluation of the retinal structures of interest.
Methods:
Our graph cut-based algorithm associates each pixel in retinal images with all other pixels by weights which we have customized based on prior anatomical information, such as proximity to the fovea, scan orientation, and nerve fiber layer brightness. Dijkstra’s algorithm utilizes this weighting matrix to find the shortest weighted paths across an image, thus effectively segmenting the retinal layers. A total of 108 macular B-scans from 10 normal adult subjects were segmented manually by one grader and automatically using our software. To estimate inter-expert-observer variability, a subset of 29 B-scans was graded manually by two experts.
Results:
We measured the average thickness of 7 retinal layers in each B-scan. We calculated the absolute value difference of the average layer thicknesses between the manual and automatic estimates and likewise between the two manual expert graders. The mean and standard deviation of these differences across all B-scans are compared in Table 1.
Conclusions:
The automatic algorithm accurately segmented 7 retinal layers, with consistent results better or equal to the observed inter-expert-variability. This automated approach may significantly reduce the resources and time necessary to conduct large-scale ophthalmic studies.
Keywords: imaging/image analysis: clinical • retina • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)