[Skip to Navigation]
Invited Commentary
February 13, 2020

Data-Driven, Feature-Agnostic Deep Learning vs Retinal Nerve Fiber Layer Thickness for the Diagnosis of Glaucoma

Author Affiliations
  • 1Department of Ophthalmology, University of Washington, Seattle
  • 2Paul Allen School of Computer Science and Engineering, University of Washington, Seattle
  • 3eScience Institute, University of Washington, Seattle
JAMA Ophthalmol. 2020;138(4):339-340. doi:10.1001/jamaophthalmol.2019.6143

In this issue of JAMA Ophthalmology, Thompson et al1 report that a deep learning (DL) model using unsegmented spectral-domain optical coherence tomography (SD-OCT) scans to detect glaucoma performs better than retinal nerve fiber layer (RNFL) thickness parameters extracted by automated segmentation. The authors used data from the Duke Glaucoma Repository, which included 20 806 SD-OCT images from 1154 eyes of 635 individuals. A convolutional neural network DL model was trained on unsegmented, raw SD-OCT peripapillary B-scan images in a fully data-driven manner. This DL model was compared with conventional RNFL thickness measurements in its ability to discriminate glaucomatous from control eyes based on the area under the receiver operating curve (AUC) and prespecified sensitivity cutoffs of 80% and 95%. Testing of the model was stratified by preperimetric glaucoma and for mild, moderate, and severe glaucoma. The AUC for the DL model (0.96) on the entire test data was significantly higher than the AUC for the global RNFL thickness–based model (0.87) and than the AUCs of each of the sectoral RNFL thickness models (ranging from 0.62 for the temporal sector to 0.87 for the inferior temporal sector). The DL model did well for preperimetric, mild, and moderate glaucoma. However, for the severe glaucoma subset, the difference in the performance of the DL model and the global RNFL thickness–based model was not statistically significant.

Add or change institution
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words