Deep learning and optical coherence tomography in glaucoma: Bridging the diagnostic gap on structural imaging

Abstract

Glaucoma is a leading cause of progressive blindness and visual impairment worldwide. Microstructural evidence of glaucomatous damage to the optic nerve head and associated tissues can be visualized using optical coherence tomography (OCT). In recent years, development of novel deep learning (DL) algorithms has led to innovative advances and improvements in automated detection of glaucomatous damage and progression on OCT imaging. DL algorithms have also been trained utilizing OCT data to improve detection of glaucomatous damage on fundus photography, thus improving the potential utility of color photos which can be more easily collected in a wider range of clinical and screening settings. This review highlights ten years of contributions to glaucoma detection through advances in deep learning models trained utilizing OCT structural data and posits future directions for translation of these discoveries into the field of aging and the basic sciences.

Keywords: aging; artificial intelligence; basic sciences; deep learning; glaucoma; optical coherence tomography.

Figure 1

Flow diagram of articles reviewed.

Figure 2

Comparison of Ground truth for deep learning algorithms. Studies were divided into four categories for ground truth, including clinical diagnosis, clinical reference, manual grading, and automated grading. (A) Comparison of all (far left) or sub-categorized (remaining) studies training deep learning algorithms for OCT of the optic nerve, RNFL, or macula ( Supplementary Table 1 ), (B) Comparison of all (far left) or sub-categorized (remaining) studies training deep learning algorithms for OCT of anterior chamber anatomy ( Supplementary Table 2 ), and (C) Comparison across all studies training deep learning algorithms for OCT – Photo pairs ( Supplementary Table 3 ).

Figure 3

Comparison of algorithm output for deep learning algorithms. Studies were divided into categories based on output parameters for (A) studies examining OCT of the optic nerve, RNFL, or macula ( Supplementary Table 1 ), (B) OCT of anterior chamber anatomy ( Supplementary Table 2 ), and (C) OCT – Photo pairs ( Supplementary Table 3 ).

Figure 4

Demographic reporting of gender and race/ethnicity in algorithm testing and training. Studies for training and testing of deep learning algorithms applied to: (A) OCT of the optic nerve, RNFL, or macular ( Supplementary Table 1 ), (B) OCT of the anterior chamber ( Supplementary Table 2 ), and (C) OCT-Photo pairs ( Supplementary Table 3 ) were categorized by reporting of gender and race/ethnicity (left). Studies reporting race/ethnicity were further categorized by the number of race/ethnicities compared and/or reported.

Figure 5

Races and ethnicities represented in algorithm testing and training. For studies from Figure 4 that reported race/ethnicity, we graphed the number studies reporting data for specific races/ethnicities. As in Figures 2 , 4 , we segregated studies that trained and tested deep learning algorithms for: (A) OCT of the optic nerve, RNFL, or macular ( Supplementary Table 1 ), (B) OCT of the anterior chamber ( Supplementary Table 2 ), and (C) OCT-Photo pairs ( Supplementary Table 3 ). For ease of comparison, color coding for individual races/ethnicities is consistent in A-C.