Diagnostic performance of automated circumpapillary retinal nerve fiber layer (RNFL) segmentation with high-resolution swept-source optical coherence tomography in Glaucoma

Abstract

Purpose : Glaucoma is a progressive optic neuropathy that leads to loss of retinal ganglion cells and thinning of retinal nerve fiber layer (RNFL), causes visual field loss and ultimately irreversible blindness. Thinning of RNFL can be quantified by performing objective thickness measurement along circumpapillary RNFL using optical coherence tomography (OCT), for glaucoma diagnostic and monitoring purposes. However, manual measurement is tedious. We evaluate the diagnostic performance of automated circumpapillary RNFL segmentation in swept-source OCT (SS-OCT) images.

Methods : SS-OCT volumes of the optic nerve head (ONH) comprising 250 glaucomatous and 75 normal eyes were collected using a Plex Elite 9000 (Carl Zeiss Meditec, USA) OCT system. Each acquired volumetric scan was processed to reconstruct a single circumpapillary cross-sectional image along a 3.46mm diameter circle centred at the ONH. A trained deep convolutional neural network using a U-Net architecture was then used to segment the circumpapillary RNFL from the cross-sectional images. A global RNFL thickness value was tabulated for each eye by averaging the obtained RNFL profile. Receiver operating characteristic (ROC) analysis was used to assess the overall diagnostic performance.

Results : Mean severity of the glaucoma eyes in our study was -3.64 ± 0.20 dB. For the glaucoma eyes, the mean thickness of the segmented RNFL was 84.97 ± 0.74 um, which was significantly different (P < .001) from that of the normal eyes (104.24 ± 1.21 um). As shown in Fig 1, our automated RNFL segmentation achieved an area under the ROC curve (AUC) of 0.91 (95% CI: 0.86 to 0.94). Example segmentation results are shown in Fig 2.

Conclusions : The results show a high diagnostic accuracy for circumpapillary RNFL segmentation for glaucoma detection, indicating the potentiality of automatic segmentation of circumpapillary RNFL in both healthy and glaucoma cases for detecting glaucoma.

This is a 2021 ARVO Annual Meeting abstract.

 

ROC curve for Circumpapillary RNFL thickness (N = 325)

 

RNFL segmentation overlay for glaucomatous (Left) and healthy (Right) eyes.