Quality of Automatic Segmentation of 3D Macular Retinal Imaging With a HD-OCT in Patients With Glaucoma and Healthy Volunteers

Abstract

 

Purpose:

 

To evaluate the reliability of layer-segmentation of three-dimensional rapid scanning, spectral-domain high-definition optical coherence tomography (HD-OCT) using a newly developed software algorithm.

 

Methods:

 

20 by 20 degree raster-scanning macular imaging is performed using a HD-OCT with an axial resolution of ~6 µm and a resolution of 512*128*1024 voxels. A novel software was developed to obtain an automated segmentation of the neuroretinal layers according to their histological architecture. 13 eyes of 13 healthy volunteers and 10 eyes of 10 patients with glaucomatous optic neuropathy were evaluated using HD-OCT retinal imaging. The following borders were analyzed: the inner limiting membrane (ILM), the retinal pigment epithelium (RPE) and the posterior borders of the following layers: retinal nerve fiber layer (RNFL), retinal ganglion cell layer (RGCL), inner plexiform layer (IPL), inner nuclear layer (INL), outer plexiform layer (OPL), photoreceptor layer (PRL, this included the outer nuclear layer and the inner segments). Automatically detected borders of retinal layers were subjectively compared to the visible layer borders by 2 experts independently to detect algorithm errors. A sample of 9 predefined b-scans per eye was analyzed. Criteria for border detection failure were: 1) border wandering to different retinal structure or obvious deviation of automatically detected border by >20 µm together with 2) a length of deviation >5% of the b-scan continuously or >20% cumulated. For a macular 3D HR-OCT scan quality of border was labeled as good, sufficient or bad if no more than 1, 1 to 3, or more than 3 individual b-scans were deemed border detection failures.

 

Results:

 

 

 

Conclusions:

 

We demonstrate high reliability of macular HD-OCT segmentation for most of the neuroretinal layer borders, when compared to 2 experts' subjective judgements. Automatic HD-OCT segmentation reliably follows the visible borders of retinal layer structures.

 

Keywords: imaging/image analysis: clinical