Fast Automated Segmentation of Fluid Spaces in Fourier-Domain OCT Images of the Retina

Abstract

 

Purpose:

 

To date the assessment of optical coherence tomography (OCT) images of the retina requires considerable manual interaction. We evaluated the ability of a novel fully automated technique to segment cystoid and subretinal fluid spaces in Fourier-domain (FD) OCT images.

 

Methods:

 

Regions of hyporeflectivity (e.g. cyst, vitreous space or retinal tissue) were first segmented from the original FD OCT image (zoom-in view in Fig. A) by a fast globally convex segmentation algorithm without necessitating initialisation. Those regions detected outside the retina were then automatically removed by considering their relative locations to the inner limiting membrane and retinal pigment epithelium estimated by curve fitting technique. Falsely detected regions within the retina were also automatically eliminated by using differentiating features (e.g. size, intensity and shape).

 

Results:

 

20 B-scan images (one from each volume scan of the Heidelberg Spectralis OCT) were studied. Mean processing time for a B-scan was 5.3 seconds. The performance of the program was validated by comparing the areas automatically detected (Fig. B) with those manually delineated by medical retina specialists (Fig. C). t-test showed no statistical difference (p=0.2).

 

Conclusions:

 

A fast, fully automated segmentation approach showed promising results for segmentation of cystoid and subretinal fluid spaces in OCT images and has the potential to aid treatment and management of eye disease.  

 

Keywords: imaging/image analysis: clinical • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • retina