Predicting Future Visual Field Status from Baseline OCT Scan in Early to Moderate Glaucoma Using Deep Learning
Related Experiment Video
Updated: Nov 9, 2026
Assessing Early Stage Open-Angle Glaucoma in Patients by Isolated-Check Visual Evoked Potential
Published on: May 25, 2020
Abadh K Chaurasia1,2, Chen Wang3, Patrick W Toohey2
1Menzies Institute for Medical Research, University of Tasmania, Hobart, Australia.
Objective:
The visual field (VF) test results of many eyes with glaucoma progress despite treatment. This suggests that some eyes are either untreated or that the management of intraocular pressure does not influence the outcome. In this work, we explore whether future VF parameters can be predicted from a baseline OCT-retinal nerve fiber layer (OCT-RNFL) scan using a deep learning model.
Design:
Retrospective multicenter cohort study with external validation.
Participants:
A total of 1792 eyes from 1610 patients were used for model development. External validation was performed on 150 eyes from a second center (Zeiss Cirrus OCT) and 280 eyes from a third center (Heidelberg Spectralis OCT).
Methods:
A Vision Transformer (ViT)-based regression model was trained on baseline OCT-RNFL scans to predict 3 key VF indices (follow-up interval: 4.74 ± 2.59 years). Model performance was evaluated using regression metrics with 95% confidence intervals. Saliency maps were generated to identify OCT-RNFL regions for making decisions in predicting functional visual status.
Main Outcome Measures:
Future VF indices, including mean deviation (MD), pattern standard deviation (PSD), and Visual Field Index (VFI), were predicted and evaluated using mean absolute error (MAE) and root mean squared error (RMSE).
Results:
The model achieved MAE of 3.23 (2.91-3.56), 1.76 (1.61-1.93), and 1.14 (1.03-1.27) for future VFI, MD, and PSD in internal validation, with comparable performance in external data sets (Zeiss OCT: 3.19 [2.73-3.66], 1.92 [1.69-2.16], and 1.11 [0.95-1.29]; Heidelberg OCT: 4.14 [3.82-4.46], 2.71 [2.51-2.91], and 1.37 [1.21-1.55]). The corresponding RMSE values reflect similar trends across the internal and external data sets. Saliency maps revealed that the inner and outer RNFL layers were key structures in driving the model's predictions.
Conclusions:
Our ViT-based regression model effectively predicts key future VF indices objectively from a single OCT-RNFL scan, with strong performance across 2 OCT devices, offering a novel tool for predicting functional visual status.
Financial Disclosures:
Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.
