TOPLINE:
A deep-learning model trained on optical coherence tomography scans of the optic nerve head reliably distinguished among various types of optic nerve damage, such as glaucoma, non-arteritic anterior ischemic optic neuropathy (NAION), and optic neuritis, although classifying optic neuritis proved the most challenging for the algorithm.
METHODOLOGY:
- Researchers conducted a cross-sectional study using data from multiple clinical trials and referral centers to determine whether a three-dimensional deep-learning model trained on optical coherence tomography scans of the optic nerve head can reliably distinguish optic atrophy in glaucoma, NAION, and optic neuritis as well as healthy eyes.
- The analysis included 7014 scans from 1382 eyes of patients with glaucoma (n = 113), NAION (n = 391), optic neuritis (n = 163) and control individuals (n = 715).
- The model was trained in three different settings, with one assessing the full optical coherence tomography volume, another focusing on the peripapillary region, and the third considering only the optic nerve head.
TAKEAWAY:
- The model analyzing the full optical coherence tomography volume achieved an overall accuracy of 88.9% with a macro-average area under the curve of 0.977; the F1 score, an indicator of the accuracy of the model, was 0.94, 0.87, 0.78, and 0.91 for glaucoma, NAION, optic neuritis, and healthy eyes, respectively.
- The other models achieved greater than 85% overall accuracy and a macro-average area under the curve of around 0.97, indicating reasonable or very good overall accuracy and discriminative capability, respectively.
- Optic neuritis was the hardest to classify across all three settings, with F1 scores between 0.71 and 0.78; some cases were misclassified as NAION or healthy; further analysis showed eyes with thinner layers of nerve fibers were labeled as NAION, whereas those with near-normal fiber thickness were labeled healthy.
- Activation maps revealed distinct structural signatures in the retinal nerve fiber layer, the retinal pigment epithelium, and other regions for each condition.
IN PRACTICE:
“Our findings highlight that optic nerve diseases exhibit distinct patterns of atrophy, which could support retrospective diagnostic efforts in cases lacking formal diagnoses. While many clinicians struggle to diagnose solely based on [retinal nerve fiber layer] thickness patterns because of their subtlety and the overlap between conditions, this tool may assist in identifying disease-specific signatures,” the researchers reported.
This new study “highlights the potential for not only distinguishing healthy vs diseased eyes but further increasing granularity by discerning individual pathology,” experts wrote in an editorial accompanying the journal article. “With further refinement, this effort could result in a global, device-agnostic, multisystem classification algorithm that could help identify a variety of optic neuropathies,” they added.
SOURCE:
The study was led by David Szanto, Icahn School of Medicine at Mount Sinai in New York City. It was published online on August 21 in JAMA Ophthalmology.
LIMITATIONS:
The study had relatively few scans for each disorder, particularly for optic neuritis. The data sources were geographically narrow, with glaucoma cases sourced exclusively from the University of Iowa Health System and optic neuritis cases from a single neuro-ophthalmology clinic in New York City. The disease groups were not matched for degree of visual dysfunction.
DISCLOSURES:
This study was supported in part by the New York Eye and Ear Infirmary Foundation, National Eye Institute, Research to Prevent Blindness, and other sources. Some authors reported receiving grants from and having patents and other ties with various sources.
This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.