Retinal Surface Contour is Predictive of Fast Glaucoma Progression with Deep Learning

Abstract

Purpose : We aim to further study if retinal surface contour (RSC) represented by inner limiting membrane at the optic nerve head is predictive of visual field (VF) progression using a deep learning model.

Methods : We selected reliable optical coherence tomography (OCT) scans from patients with at least 5 reliable 24-2 VFs over at least 4 years. We calculated 4 different VF progression outcomes: (1) mean deviation (MD) progression: MD slope < 0 and p-value < 0.05; (2) VF index (VFI) progression: VFI slope < 0 and p-value < 0.05; (3) total deviation (TD) pointwise progression: at least 3 TD locations with TD slope ≤ -1 dB/year and p-value < 0.05; (4) MD fast progression: MD slope ≤ -1 dB/year and p-value < 0.05. For each progression outcome, we built three deep learning models based on EfficientNet using the following inputs: (1) RNFLT map; (2) RSC map (Figure. 1); (3) RNFLT plus RSC maps. Our models were trained using two-thirds of the dataset and then tested using the remaining one-third with patient-level separation. T-test with bootstrapping sampling was used to compare the performance of different models measured by the area under the receiver operating characteristic curve (AUC).

Results : 16,155 OCT scans from 5,166 eyes of 3,115 patients with progression outcomes were included with 10,963 and 5,192 for training and testing, respectively. The average age and MD at baseline were 61.6 ± 13.0 years and -3.0 ± 4.1 dB, respectively. The progression percentages for MD progression, VFI progression, TD pointwise progression, and MD fast progression were 9.7%, 11.1%, 11.3%, and 2.6% with a median follow-time of 6.6 years, respectively. The AUCs of using RNFLT map, RSC map, and their combination for the four progression outcomes are as follows: (1) MD progression: 0.79, 0.69, and 0.81; (2) VFI progression: 0.81, 0.71, and 0.83; (3) TD pointwise progression: 0.84, 0.77 and 0.85; (4) MD fast progression: 0.76, 0.83 and 0.89 (Figure. 2). Combining RNFLT and RSC maps consistently performed better at predicting VF progression than using each of them alone. Although the RNFLT map generally outperformed the RSC map for progression prediction, the RSC map better predicted MD fast progression than the RNFLT map (AUCs: 0.83 versus 0.76, p < 0.001).

Conclusions : Combining RSC with RNFLT map can improve glaucoma progression prediction. RSC map is a better predictor of fast progression than the RNFLT map.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

Figure 1

 

Figure 2