Abstract
Introduction: With normal aging, various neurobiological alterations (NBAs) occur that cause subtle visuo-perceptual deficits. These deficits are generally greater for complex visual information. Identifying which NBA causes these deficits is difficult because such deficits are subtle and may be the result of various NBAs emerging more or less simultaneously. The goal of our study was to simulate various NBA using an artificial neural network to identify which one correlates best with psychophysical human aging findings. Methods: An artificial neural network learned to perform an orientation-identification task for simple, luminance-defined (first-order) and complex, texture-defined (second-order) stimuli. Habak & Faubert (2000) demonstrated that orientation-identification thresholds for a group of elderly observers (X=70.1 years) increased by 29 % for first-order stimuli and 77 % for second-order stimuli compared to young adults (X=23 years). Once the artificial neural network learned the tasks, various NBAs were simulated. Results: Simulations showed that several NBA hypotheses including synapse loss, myelin sheath degradation and degeneration of intra-cortical inhibition correlated with the psychophysical results for the two age groups tested by Habak & Faubert (2000). However, these hypotheses did not predict the same relative loss in performances for other age groups, such as the middle-aged and very elderly observers. Conclusion: The artificial neural networks were able to predict relative threshold increase reflecting different types of physiological changes during the aging process for two age groups. However, psychophysical data for other age groups are required to differentiate between NBAs.
This work was supported by CIHR and NSERC/Essilor Industrial Research Chair