Related Experiment Video
Updated: Aug 24, 2025
Assessing Early Stage Open-Angle Glaucoma in Patients by Isolated-Check Visual Evoked Potential
Published on: May 25, 2020
Prior optic neuritis detection on peripapillary ring scans using deep learning
Seyedamirhosein Motamedi1, Sunil Kumar Yadav1,2, Rachel C Kenney3,4
1Experimental and Clinical Research Center, a cooperation between the Max Delbrück Center for Molecular Medicine in the Helmholtz Association and Charité Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt-Universität zu Berlin, Berlin, Germany.
Deep learning models can effectively identify prior optic neuritis (ON) in multiple sclerosis (MS) patients using retinal scans. This AI approach shows promise in improving the diagnosis of demyelinating events.
Area of Science:
- Ophthalmology
- Neurology
- Artificial Intelligence
Background:
- Multiple sclerosis diagnosis requires evidence of demyelination disseminated in time and space.
- Peripapillary retinal nerve fiber layer (pRNFL) thickness via optical coherence tomography (OCT) may indicate prior acute optic neuritis (ON), a common MS symptom.
Purpose of the Study:
- To evaluate a deep learning (DL) network's ability to differentiate eyes with a history of ON from healthy control (HC) eyes.
- To compare the DL network's performance against pRNFL thickness measurements.
Main Methods:
- A dilated residual convolutional neural network was trained and validated on 1033 OCT scans from HC eyes and 510 scans from eyes with prior ON.
- An independent dataset from a second center was used for external validation.
- Performance was assessed using receiver operating characteristic curve analyses and area under the curve (AUC).
Main Results:
- The DL network achieved an AUC of 0.86 in recognizing ON eyes, outperforming pRNFL thickness (AUC 0.77).
- On the independent dataset, the DL network achieved an AUC of 0.90, compared to pRNFL's AUC of 0.84.
Conclusions:
- Deep learning-based classification of prior ON is feasible.
- DL models have the potential to surpass traditional thickness-based methods for classifying eyes with and without a history of ON.
