Lazaridis, G;
Lorenzi, M;
Mohamed-Noriega, J;
Aguilar-Munoa, S;
Suzuki, K;
Nomoto, H;
Ourselin, S;
... United Kingdom Glaucoma Treatment Study Investigators; + view all
(2021)
OCT Signal Enhancement with Deep Learning.
Ophthalmology Glaucoma
, 4
(3)
pp. 295-304.
10.1016/j.ogla.2020.10.008.
Preview |
Text
Manuscript.pdf - Accepted Version Download (1MB) | Preview |
Abstract
PURPOSE: To establish whether deep learning methods are able to improve the signal-to-noise ratio of time-domain (TD) optical coherence tomography (OCT) images to approach that of spectral-domain (SD) OCT. DESIGN: Method agreement study and progression-detection in a randomized, double-masked, placebo-controlled, multi-centre trial for open-angle glaucoma (OAG) [UK Glaucoma Treatment Study (UKGTS)]. PARTICIPANTS: Cohort for training and validation: 77 stable OAG participants with TDOCT and SDOCT imaging at up to 11 visits within 3 months. Cohort for testing: 284 newly-diagnosed OAG patients with TDOCT from a cohort of 516 recruited at 10 UK centres between 2007 and 2010. METHODS: An ensemble of generative adversarial networks (GANs) was trained on TDOCT and SDOCT image pairs from the training dataset and applied to TDOCT images from the testing dataset. TDOCT were converted to synthesized SDOCT images and segmented via Bayesian fusion on the output of the GANs. MAIN OUTCOME MEASURES: 1) Bland-Altman analysis to assess agreement between TDOCT and synthesized SDOCT average retinal nerve fibre layer thickness (RNFLT) measurements and the SDOCT RNFLT. 2) Analysis of the distribution of the rates of RNFLT change in TDOCT and synthesized SDOCT in the two treatments arms of the UKGTS was compared. A Cox model for predictors of time-to-incident VF progression was computed with the TDOCT and the synthesized SDOCT. RESULTS: The 95% limits of agreement between TDOCT and SDOCT were [26.64, -22.95], between synthesized SDOCT and SDOCT were [8.11, -6.73], and between SDOCT and SDOCT were [4.16, -4.04]. The mean difference in the rate of RNFL change between UKGTS treatment and placebo arms with TDOCT was 0.24 (p=0.11) and with synthesized SDOCT was 0.43 (p=0.0017). The hazard ratio for RNFLT slope in Cox regression modeling for time to incident VF progression was 1.09 (95% CI 1.02 to 1.21) (p=0.035) for TDOCT and 1.24 (95% CI 1.08 to 1.39) (p=0.011) for synthesized SDOCT. CONCLUSIONS: Image enhancement significantly improved the agreement of TDOCT RNFLT measurements with SDOCT RNFLT measurements. The difference, and its significance, in rates of RNFLT change in the UKGTS treatment arms was enhanced and RNFLT change became a stronger predictor of VF progression.
Archive Staff Only
 |
View Item |
