Purchase this article with an account.
Akram Belghith, Felipe A Medeiros, Christopher Bowd, Robert N Weinreb, Tu Zhuowen, Linda M Zangwill; A novel texture-based OCT enface image to detect and monitor glaucoma. Invest. Ophthalmol. Vis. Sci. 201657(12):.
Download citation file:
© 2017 Association for Research in Vision and Ophthalmology.
To compare a novel texture-based enface image, the standard intensity-based enface image and retinal nerve fiber layer (RNFL) thickness for open-angle glaucoma (OAG) detection and to assess the ability of deep-learning convolution autoencoders to identify areas of interest from enface images and RNFL thickness maps for monitoring glaucoma.
Cross-sectional data were collected from 57 eyes of 29 healthy subjects and 181 eyes of 96 OAG patients. Longitudinal data were collected from 95 eyes (35 progressing OAG eyes and 60 healthy eyes) of 63 participants followed for an average of 3.7 years. SD-OCT optic nerve head volume and RNFL circular scans were acquired. RNFL thickness maps were generated using San Diego Automated Layer Segmentation Algorithm (SALSA). 2 OCT enface images based on the 1) average intensity and 2) tissue texture were generated from 70-μm slabs following the Inner limiting membrane. For each eye, deep-learning autoencoders were applied on the enface images and RNFL thickness map to automatically identify individualized areas of interest for OAG monitoring based on image feature ranking. Area under the receiver operating curves (AUC) was used to compare diagnostic accuracy of the circumpapillary (cp)RNFL thickness, RNFL thickness map, texture enface and intensity enface image. An eye was defined as progressing when structural loss was significantly different from zero and faster than the 5th percentile of the healthy group.
For the cross-sectional and longitudinal group, healthy subjects were significantly younger than OAG subjects. The novel texture enface images and the RNFL thickness map performed significantly better than standard cpRNFL and intensity enface images for discriminating between OAG eyes (mean (± SD) visual field MD =-3.07 dB (±4.35)) and healthy eyes (age-adjusted AUC of 0.94, 0.93, 0.91 and 0.86, respectively). Among the 35 OAG progressing eyes (mean visual field MD =-7.7 dB (±4.8)), 31 (89%) eyes were identified as progressing by the RNFL thickness map, 30 (86%) eyes by texture enface image, 25 (71%) eyes by cpRNFL and 14 (40%) eyes by intensity enface image.
Novel texture enface images and RNFL thickness maps were significantly better than standard intensity enface images and cpRNFL thickness for discriminating between healthy and OAG eyes and for detecting progression. Deep learning methods show promise for individualizing monitoring of glaucoma.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
This PDF is available to Subscribers Only