[Paper Review] Deep Learning and Computer Vision for Glaucoma Detection: A Review

Mona Ashtari-Majlan, Mohammad Mahdi Dehshibi+1 more|arXiv (Cornell University)|Jul 31, 2023

Retinal Imaging and Analysis4 citations

TL;DR

This paper reviews deep learning and computer vision approaches for automated glaucoma detection using fundus, OCT, and visual field images, emphasizing architectural paradigms, performance benchmarks, and open challenges. It identifies critical gaps in generalizability, uncertainty estimation, and multimodal integration, while curating datasets and advocating for reproducible, clinically translatable AI solutions.

ABSTRACT

Glaucoma is the leading cause of irreversible blindness worldwide and poses significant diagnostic challenges due to its reliance on subjective evaluation. However, recent advances in computer vision and deep learning have demonstrated the potential for automated assessment. In this paper, we survey recent studies on AI-based glaucoma diagnosis using fundus, optical coherence tomography, and visual field images, with a particular emphasis on deep learning-based methods. We provide an updated taxonomy that organizes methods into architectural paradigms and includes links to available source code to enhance the reproducibility of the methods. Through rigorous benchmarking on widely-used public datasets, we reveal performance gaps in generalizability, uncertainty estimation, and multimodal integration. Additionally, our survey curates key datasets while highlighting limitations such as scale, labeling inconsistencies, and bias. We outline open research challenges and detail promising directions for future studies. This survey is expected to be useful for both AI researchers seeking to translate advances into practice and ophthalmologists aiming to improve clinical workflows and diagnosis using the latest AI outcomes.

Motivation & Objective

• To provide a systematic review of deep learning and computer vision techniques for automated glaucoma diagnosis from 2017 to 2023.
• To address the diagnostic challenges of glaucoma, including its asymptomatic early stage and limitations in current imaging and functional testing.
• To evaluate performance gaps in model generalizability, uncertainty estimation, and multimodal data integration across public datasets.
• To curate key datasets and highlight limitations such as scale, labeling inconsistencies, and bias in existing benchmarks.
• To outline open research challenges and future directions for translating AI advances into clinical practice for improved glaucoma management.

Proposed method

• Conducted a systematic literature search across Web of Science, PubMed, IEEE Xplore, and Google Scholar using targeted keywords and Boolean operators.
• Employed a two-phase screening process: title/abstract review followed by full-text evaluation to ensure relevance to AI-based glaucoma diagnosis.
• Categorized deep learning models into architectural paradigms including CNNs, autoencoders, attention networks, GANs, and geometric deep learning.
• Benchmarked models on widely used public datasets to assess performance, generalizability, and robustness across different imaging modalities.
• Evaluated feature extraction techniques—structural, statistical, and hybrid—used to identify glaucoma biomarkers from retinal images.
• Promoted reproducibility by linking reviewed methods to available source code repositories and advocating for open datasets and evaluation benchmarks.

Figure 1: Anatomical structures of the human eye and optic nerve relevant to glaucoma detection. [Left] Schematic views, [Right] Fundus and OCT views [ 11 ] . This figure was created using images licensed under Creative Commons.

Experimental results

Research questions

• RQ1 What are the current state-of-the-art deep learning architectures and computer vision techniques used for glaucoma detection across fundus, OCT, and visual field imaging?
• RQ2 How do existing models perform in terms of generalizability, uncertainty estimation, and multimodal integration on public datasets?
• RQ3 What are the key limitations in current datasets, including scale, labeling quality, and bias, and how do they affect model performance?
• RQ4 What are the major challenges in translating AI-based glaucoma diagnosis into real-world clinical workflows?
• RQ5 What future research directions—such as meta-learning, adversarial training, and reinforcement learning—can accelerate clinical adoption?

Key findings

• Deep learning models, particularly CNNs and attention-based architectures, demonstrate strong performance in classifying glaucoma from fundus and OCT images, with some models achieving AUCs above 0.90 on benchmark datasets.
• Significant performance gaps remain in generalizability across diverse populations and imaging devices, indicating limited robustness to domain shift.
• Uncertainty estimation in deep learning models for glaucoma remains underdeveloped, with few methods providing reliable confidence estimates for clinical decision support.
• Multimodal integration of fundus, OCT, and visual field data is still in early stages, with limited models effectively combining these data types to improve diagnostic accuracy.
• Existing datasets suffer from limitations in size, labeling inconsistency, and demographic bias, which hinder the development of robust and equitable AI systems.
• The integration of explainable AI, transfer learning, and physician-in-the-loop frameworks is emerging as a critical path toward clinical adoption and trust in AI-driven glaucoma diagnosis.