Mosaicing and Multi-Field Layer Segmentation of 3D Retinal OCT

Abstract

 

Purpose

 

First, to increase the retinal OCT coverage by creating a mosaic of OCT images from multiple spatially overlapping fields. Second, to perform a multi-field co-segmentation of intraretinal layers, assuring consistent segmentation of the fields in the overlapped areas.

 

Methods

 

A 9-field per eye acquisition was performed where a subject fixates on 9 spots in a 3×3 grid pattern. Subjects were imaged with spectral domain OCT Spectralis (768x61x496 voxels, 9.5x8.1x1.9 mm3 with voxel size of 12.41x132.22x3.87 µm3). In addition, the device acquires 2D SLO fundus image (768x768 voxels, 9.5x9.5 mm2 with voxel size of 12.41x12.41 µm2), co-registered with the OCT image. We create a mosaic by performing 2D en-face, affine alignment of the imaged fields, based on matched Speeded Up Robust Features keypoints. For the alignment, the SLO images are used as surrogate to projection OCT images due to their higher spatial resolution and pronounced texture (Figure 1). This is followed by the graph-search based multi-field co-segmentation of intraretinal layers. All 9 fields are segmented simultaneously, imposing a priori soft intrasurface-interfield constraint for each pair of overlapping fields. The constraints penalize deviations from the expected surface height differences, taken as the depth-axis shifts that produce the maximum cross-correlation of pairwise-overlapped areas.

 

Results

 

Our method was evaluated on acquisitions from 10 glaucoma patients. Qualitatively, the obtained thickness maps show no stitching artifacts, compared to pronounced stitches when the fields are segmented independently (Figure 2). Quantitatively, two ophthalmologists manually traced ILM, RNFL, GCL+IPL, and RPE layers. The average unsigned error of the automated method (4.36 ± 1.72 µm) was comparable to the average difference between the observers (5.85 ± 1.72 µm).

 

Conclusions

 

Building a mosaic of multiple SLO-OCT fields is an effective approach for increasing the coverage of imaged retina. As opposed to segmenting layers in each of the fields independently (the current state of the art), the proposed multi-field co-segmentation method obtains consistent segmentation results across the overlapped and registered areas, producing accurate and artifact-free thickness maps.

 

 

9-field SLO mosaic (left) and the corresponding composite OCT projection image (right)

 

 

9-field RNFL thickness map: without (left) and with (right) imposed layer consistency in the overlapped areas.

 

Keywords: 549 image processing • 551 imaging/image analysis: non-clinical • 688 retina