Related Experiment Video
Updated: May 2, 2026
12:23
Dynamic Visual Tests to Identify and Quantify Visual Damage and Repair Following Demyelination in Optic Neuritis Patients
Published on: April 14, 2014
14.2K
Optic Nerve Atrophy Conditions Associated With 3D Unsegmented Optical Coherence Tomography Volumes Using Deep
David Szanto1, Jui-Kai Wang2, Brian Woods3,4
1Department of Ophthalmology, Icahn School of Medicine at Mount Sinai, New York.
JAMA Ophthalmology
|August 21, 2025
Summary
A 3D deep learning model accurately differentiates optic nerve head atrophy in glaucoma, NAION, and optic neuritis using OCT scans. This automated approach aids diagnosis and clinical management of optic neuropathies.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate differentiation of optic nerve head (ONH) atrophy is crucial for diagnosing and managing conditions like glaucoma, nonarteritic anterior ischemic optic neuropathy (NAION), and optic neuritis.
- Traditional 2D assessments may miss subtle volumetric changes indicative of optic atrophy.
Purpose of the Study:
- To evaluate a 3D deep learning model's ability to distinguish optic atrophy in glaucoma, NAION, optic neuritis, and healthy eyes using unsegmented ONH optical coherence tomography (OCT) scans.
Main Methods:
- A cross-sectional study analyzed 7014 Cirrus ONH OCT scans from 1382 patients (glaucoma, NAION, optic neuritis, healthy controls).
- Three ResNet-3D-18 models were trained using 5-fold cross-validation: entire volume, peripapillary region (PPR)-only, and ONH-only.
- Performance was assessed using classification accuracy, AUC-ROC, precision, recall, and F1 scores.
Main Results:
- The entire-volume model achieved 88.9% accuracy (AUC-ROC, 0.977), with F1 scores ranging from 0.78 (optic neuritis) to 0.94 (glaucoma).
- PPR-only and ONH-only models showed slightly lower accuracies (85.9% and 87.0%, respectively).
- Optic neuritis was the most challenging to classify, often misclassified as NAION or healthy. Activation maps highlighted specific retinal layers.
Conclusions:
- Deep learning analysis of unsegmented OCT scans reliably distinguishes between different forms of optic nerve atrophy, identifying subtle, disease-specific structural patterns.
- This automated approach can support diagnostic efforts and clinical management of optic neuropathies.
- It may complement less standardized imaging modalities and subjective clinical assessments.
