Glaucoma Diagnostic Accuracy Improved Using 3D Imaging Technology

The 3-dimensional (3D) structural configuration of the central retinal vessel trunk and its branches (CRVT&B) has superior diagnostic power for glaucoma than retinal nerve fiber layer (RNFL) thickness, according to a study published in the American Journal of Ophthalmology.

Glaucoma cases (n=1639) and controls (n=2469) were recruited at the Singapore National Eye Centre, the Aravind Eye Hospital in India, and the Vilnius University Hospital Santaros Klinikos in Lithuania. All participants underwent standard spectral-domain ocular coherence tomography (SD-OCT) examinations and images were used to assess diagnostic markers for glaucoma using a convolutional neural network (CNN) approach. The researchers then compared the diagnostic accuracy of CRVT&B imaging data with the RNFL thickness.

Study participants were stratified into 8 cohorts based on recruitment center and ethnicity. Cohorts ranged in average age from 30.1 to 67.3 years and between 49% and 80% were men.

The overall 3D-CNN model was able to detect glaucoma with an accuracy of 80.3%, sensitivity of 79.5%, specificity of 80.8%, and area under the curve (AUC) of 0.84±0.02.

The overall 2-dimensional (2D) CNN model detected glaucoma with an accuracy of 84.5%, sensitivity of 86.5%, specificity of 83%, and AUC of 0.88±0.02.

The cross-sectional view from the temporal direction had the best diagnostic accuracy (AUC, 0.84) followed by the enface view (AUC, 0.82) and the cross-sectional view from the superior direction (AUC, 0.81). Combining all 3 views maximized accuracy (AUC, 0.87).

Compared with the standard biomarker for glaucoma, RNFL thickness, the AUCs for classifying glaucoma were always higher for CRVT&B than for RNFL thickness. Both the 2D-CNN (AUC, 0.90) and 3D-CNN (AUC, 0.89) outperformed RNFL thickness in cohorts one (AUC, 0.80), two (AUC, 0.78), and three (AUC, 0.74).

This study may have been limited by not including diastolic and systolic ocular perfusion pressure data, however, it remains unclear whether current computational resources are sufficient to include additional information.

These data demonstrated the potential power of using deep learning networks for diagnosing glaucoma.

Disclosure:  One study author declared affiliations with the biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures. 

Reference

Panda SK, Cheong H, Tun TA, et al. The three-dimensional structural configuration of the central retinal vessel trunk and branches as a glaucoma biomarker. Am J Ophthalmol. Published online March 2, 2022. doi:10.1016/j.ajo.2022.02.020