Abstract
Purpose :
To develop a computational model for predicting contrast thresholds of localized spatial patterns across the visual field based on a model of the eye that includes retinal ganglion cell (RGC) responses, and to apply the model towards evaluating known structure-function relationships in glaucoma.
Methods :
Our model is developed by modifying the retina-V1 (RV1) model (Bradley et al., 2014; J. Vis.), which includes a model of parvocellular (P) ganglion cell responses and predicts contrast thresholds of 43 different stimuli in the ModelFest dataset (Watson & Ahumada, 2005; J. Vis.)at the fovea. We extended the RV1 model to include magnocellular (M) ganglion cell responses because the RV1 model in its original form does not predict standard automated perimetry (SAP) thresholds well. In SAP, a stimulus is presented with an abrupt onset and offset suggesting that M cells, which are sensitive to high temporal frequency stimuli, are important for detection performance. M cell parameters in the modified retina-V1 (mRV1) model are estimated by fitting the central 30o of Hermann’s hill of vision (HoV) (Hermann et al., 2008; Acta Ophthalmol.). Different types of RGC damage were simulated, including axonal loss and dendritic field (DF) loss, based on the pathophysiology of glaucoma.
Results :
The mRV1 model predicts contrast thresholds for both the ModelFest dataset (RMSE = 1.18 dB) and for the central 30o of Hermann’s HoV (RMSE = 0.82 dB) well. Without modifying any parameters, simulated RGC damage predicts a well-known but poorly understood structure-function relationships in glaucoma: a large initial decrease in retinal nerve fiber layer (RNFL) thickness is accompanied by only a relatively small sensitivity loss (Hood & Kardon, 2007; Prog. Retin. Eye Res.). Simulated axonal and DF loss predicts the data in Hood & Kardon (2007) better than simulated axonal loss alone (RMSE = 0.23 vs. 0.26 on a linear scale of sensitivity).
Conclusions :
The mRV1 model is grounded in the known anatomy and physiology of the eye and predicts SAP thresholds well within the central 30o of visual field. Different types of RGC damage simulated with the mRV1 model may help explain known structure-function relationships in glaucoma that are currently not well understood.
This is a 2021 ARVO Annual Meeting abstract.