RNFL Maps and Deep Learning
Ophthalmology Glaucoma, November/December 2019
Wang et al. used full retinal nerve fiber layer (RNFL) thickness maps from patients with glaucoma and healthy controls to evaluate the diagnostic accuracy of four different machine learning algorithms. They found that all four models achieved similarly high diagnostic accuracies.
For this case-control study, the researchers evaluated 69 patients (93 eyes) with glaucoma and 128 healthy controls (128 eyes) from the Los Angeles Latino Eye Study (LALES). There was no significant difference in age, sex, best-corrected visual acuity, or axial length between the two groups.
RNFL maps centered on the optic nerve head were supplied to two conventional machine learning algorithms and two convolutional neural nets, one of which was a custom-made deep learning network. AUC (area under the curve) values for the four models were greater than 0.90 (range, 0.91-0.92). In contrast, the AUC for mean circumpapillary RNFL thickness was 0.76 in the same patient population.
The findings support the importance of the spatial structure of RNFL thickness map data in diagnosing glaucoma, the researchers said. They cautioned that as the study participants were from the LALES, the results may not be applicable to other ethnic populations.
The original article can be found here.