Memory-related cognitive modulation of human auditory cortex: Magnetoencephalography-based validation of a computational model

Thumbnail Image


umi-umd-5225.pdf (7.85 MB)
No. of downloads: 1349

Publication or External Link






It is well known that cognitive functions exert task-specific modulation of the response properties of human auditory cortex. However, the underlying neuronal mechanisms are not well understood yet. In this dissertation I present a novel approach for integrating 'bottom-up' (neural network modeling) and 'top-down' (experiment) methods to study the dynamics of cortical circuits correlated to shortterm memory (STM) processing that underlie the task-specific modulation of human auditory perception during performance of the delayed-match-to-sample (DMS) task. The experimental approach measures high-density magnetoencephalography (MEG) signals from human participants to investigate the modulation of human auditory evoked responses (AER) induced by the overt processing of auditory STM during task performance. To accomplish this goal, a new signal processing method based on independent component analysis (ICA) was developed for removing artifact contamination in the MEG recordings and investigating the functional neural circuits underlying the task-specific modulation of human AER. The computational approach uses a large-scale neural network model based on the electrophysiological knowledge of the involved brain regions to simulate system-level neural dynamics related to auditory object processing and performance of the corresponding tasks. Moreover, synthetic MEG and functional magnetic resonance imaging (fMRI) signals were simulated with forward models and compared to current and previous experimental findings. Consistently, both simulation and experimental results demonstrate a DMSspecific suppressive modulation of the AER and corresponding increased connectivity between the temporal auditory and frontal cognitive regions. Overall, the integrated approach illustrates how biologically-plausible neural network models of the brain can increase our understanding of brain mechanisms and their computations at multiple levels from sensory input to behavioral output with the intermediate steps defined.