Segmentation of Multiple Intra-retinal Surfaces in Volumetric SD-OCT Images of Mouse Eyes Using an Improved Iowa Reference Algorithm

Abstract

 

Purpose

 

The segmentation of intra-retinal surfaces in human volumetric SD-OCT images is a well-studied problem. Here, we adapt an existing graph-theoretic approach to segment 10 intra-retinal surfaces in mouse SD-OCT volumes and assess the reproducibility of the method.

 

Methods

 

Repeat scans obtained on the same day from 18 mice on a Bioptigen SD-OCT scanner was used in this study. The volumes were obtained from a region 1.4mm x 1.4mm x 2mm and had 400 x 400 x 1024 voxels. The optic disc was located manually in projection images and used to center the scans in order to avoid misalignment errors. Next, 10 intra-retinal layers (Fig. 1A) were segmented in each volume. The Iowa Reference Algorithm was adapted and improved to tackle the different anatomy. The average thickness of the 9 layers as well as the total retinal thickness was computed in 8 sectors (as depicted in Fig. 1b). The diameters of the 3 circles used are 0.2mm, 0.6mm and 1.2mm and were chosen to avoid the optic disc and peripheral regions. The difference in mean thickness (in microns) computed in the local regions between repeat scans was used as the measure of reproducibility.

 

Results

 

The overall mean difference between repeats in all ten layers was 1.19 ± 0.85 µm. The mean difference in the thickness of the retinal nerve fiber layer (RNFL), inner plexiform layer (IPL) and total retinal thickness was 2.59 ± 1.55 µm,1.14 ± 0.84 µm and 3.53 ± 2.40 µm, respectively. The mean NFL+IPL and total retinal thickness are closely correlated with correlation coefficients of 0.82 and 0.67, respectively.

 

Conclusions

 

The reproducibility of the mathematically enhanced Iowa Reference Algorithm shows sub-pixel accuracy and the mean thickness measurements are closely correlated in standard SD mouse OCT. We expect the use of a more reliable registration method and developments in the algorithm to further improve the reproducibility. Automated 3D segmentation of retinal mouse OCT has potential for precise quantification of traits and measurement of intervention effects.

 

 

Fig 1. (A) The 9 intra-retinal surfaces segmented. (B) The color-coded map of the 8 regions within which the mean layer thickness was computed, showing the mean difference computed (in μm) over all 10 layers.

 

 

Fig. 2. Scatter plots of the (A) mean thickness of the NFL+IPL and (B) total retinal thickness measured in the repeat scans.

 

Keywords: 549 image processing • 688 retina • 498 diabetes