Thirty relapsing-remitting MS patients on glatiramer acetate were enrolled in a 1-year study. OCT and SDMT were measured at baseline and Year-1. A single hidden-layer feedforward ANN is constructed with 4 input and one output nodes. The inputs included the patient´s age, thickness of the retinal nerve fiber layer (RNFL), ganglion cell layer and inner plexiform layer. The output included the SDMT. The ANN underwent supervised learning using Bayesian regularization. The data consisted of 53 time-points and 15% of this data was reserved for testing the network to ensure adequate generalization.

SDMT and RNFL thickness were significantly correlated at baseline (r=0.427, p=0.026) and Year-1 (r=0.529, p=0.005). An optimized ANN with 7 parameters was designed. The accuracy of this network on the testing (and training) sets was 72.7% (83.5%). The model´s output was found to be statistically equivalent to the actual SDMT score proving that the estimation is accurate with t-test p=0.173 (p=0.989). The model converged independently of disease and therapy duration.