Segmentation of the Retinal Vasculature within Spectral-Domain Optical Coherence Tomography Volumes of Mice

Automated approaches for the segmentation of the retinal vessels are helpful for longitudinal studies of mice using spectral-domain optical coherence tomography (SD-OCT). In the SD-OCT volumes of human eyes, the retinal vasculature can be readily visualized by creating a projected average intensity image in the depth direction. The created projection images can then be segmented using standard approaches. However, in the SD-OCT volumes of mouse eyes, the creation of projec- tion images from the entire volume typically results in very poor images of the vasculature. The purpose of this work is to present and evaluate three machine-learning approaches, namely baseline, single-projection, and all-layers approaches, for the automated segmentation of retinal ves- sels within SD-OCT volumes of mice. Twenty SD-OCT volumes (400 ◊ 400 ◊ 1024 voxels) from the right eyes of twenty mice were obtained using a Bioptigen SD-OCT machine (Morrisville, NC) to evaluate our methods. The area under the curve (AUC) for the receiver operating characteristic (ROC) curves of the all-layers approach, 0.93, was signifi- cantly larger than the AUC for the single-projection (0.91) and baseline (0.88) approach with p < 0.05.