In this commentary in the journal JAMA Ophthalmology, Christine A. Petersen, MD, Parmita Mehta, MS, and Aaron Y. Lee, MD, MSCI review a recent article published in the same journal, "Assessment of a segmentation-free deep learning algorithm for diagnosing glaucoma from optical coherence tomography scans."
The commentary addresses why a deep learning approach was more successful at detecting glaucoma using spectral domain optical coherence tomography scans than the more traditional strategy of performing automated segmentation to analyze retinal nerve fiber layer parameters.
The authors note several strengths of the study, such as the fact that the glaucoma cases were stratified by severity. Accurate early detection of glaucoma is critical because it allows for initiation of therapeutic interventions to slow disease progression at an earlier stage, and the deep learning model was better at detecting glaucoma in its earliest stages. Using the feature-agnostic deep learning approach also led to the identification other retinal layers (in addition to the RNFL) that may be relevant in the pathophysiology of glaucoma, which the authors note may help researchers generate new hypotheses and provide novel insights into the disease.
Petersen CA, Mehta P, Lee AY. Data-Driven, Feature-Agnostic Deep Learning vs Retinal Nerve Fiber Layer Thickness for the Diagnosis of Glaucoma. JAMA Ophthalmol. 2020 Feb 13. doi: 10.1001/jamaophthalmol.2019.6143. [Epub ahead of print]