Our laboratory studies the relationship between what is observed in functional neuroimaging studies and the underlying neural dynamics. To do this, we had previously constructed a large-scale computer model of neuronal dynamics that performs a visual object-matching task similar to those designed for PET/fMRI studies (reviewed in Horwitz & Husain 2007). We extended the model so that it could also simulate auditory processing, thus allowing us to investigate the neural basis of auditory object processing in the cerebral cortex. This model relates neuronal dynamics of cortical processing of auditory spectrotemporal patterns to fMRI data.? ? Environmentally relevant auditory stimuli are often composed of long-duration tonal patterns (e.g., multisyllabic words, short sentences). Manipulation of those patterns by the brain requires working memory to temporarily store the pattern segments and integrate them into a percept. To understand the neural basis of this, we extended the model of auditory recognition of short-duration tonal patterns described above. A memory buffer and a gating module were added. The memory buffer increased the storage capacity; the gating module distributed the segments of the input pattern to separate locations of the memory buffer, allowing a subsequent comparison of the stored segments against the segments of a second pattern. Simulations show that the extended model performs match and mismatch of sequences of long-duration tonal patterns. We conducted an fMRI experiment using the same stimuli as employed in the simulations and found areas in the prefrontal cortex that are likely candidate brain areas for the new modules of the extended model. To understand the mechanisms of perceptual decision-making, we extended the model by incorporating a three-layer decision-making model and simulated an auditory delayed match-to-sample (DMS) task. The models simulated response times and the different patterns of neural responses (transient, sustained, increasing) are consistent with experimental data and the simulated neurophysiological activity provides insights into the neural interactions from perception to action in the auditory DMS task (Wen et al. 2008).? ? We also used this auditory neural model and fMRI to investigate the neural mechanisms responsible for tinnitus. We hypothesized that a network of brain regions from auditory processing areas to emotional processing areas contributes to and modulates tinnitus. A goal of the experiments and modeling is to contrast the neural differences between participants with sensorineural hearing loss and tinnitus from those with sensorineural hearing loss but without tinnitus. Preliminary fMRI data analysis suggests that the brain activation patterns of tinnitus subjects while listening to music or actively discriminating simple sounds differ from those of subjects from the control groups. We also collected diffusion-weighted MRI data to analyze the organization of white matter tracts in the brain. Results showed a difference between subjects with normal hearing and those with hearing loss (with and without tinnitus) in the left hemisphere white matter connecting auditory cortex to other brain regions, indicating that even though the pathology of hearing loss may be caused by peripheral damage in the ear, neural structural organization in the central nervous system also can be affected. Finally, preliminary results from simulating different possible neural mechanisms underlying tinnitus generation using our large-scale neural network model suggest that an increase in excitability and a decrease in the auditory cortex input threshold is the likeliest mechanism of tinnitus (Husain 2007).? ? FMRI data can be used to assess how different brain regions interact during the performance of cognitive tasks. The quantities that characterize these interactions are called functional or effective connectivity, but their neurobiological substrates are, however, uncertain. Functional connectivity is computed as the correlation between interregional activities; effective connectivity investigates the influence that brain regions exert on one another. Structural equation modeling (SEM) has been the main approach to examine effective connectivity. We proposed a method that, given a set of regions, performs partial correlation analysis. This method provides an approach to effective connectivity that is data driven in that it does not require prior information regarding knowing the anatomical connections. To demonstrate its practical, we reanalyzed data previously published, and as well, used simulated fMRI data from our large-scale neural model, and showed that partial correlation analysis can hint at which effective connections are structuring the interactions (Marrelec et al. 2007, 2008). We also used the simulated data to investigate other functional and effective connectivity methods (Kim & Horwitz 2008).? ? Many brain disorders result from alterations in the strength of anatomical connectivity between brain regions. These disorders have been studied using functional neuroimaging and techniques that evaluate interregional effective connectivity. For example, using PET data, we showed that there is altered effective connectivity among emotion-related brain regions in Alzheimer disease patients during a short-term memory task (Rosenbaum et al. 2008). In another study, we investigated how the diminution of brain anatomical connectivity can be revealed by the change of brain functional interactions. To do this, we applied SEM analysis to simulated fMRI data from our neural network model. Two subject groups were simulated: a normal subject group and a patient group for whom the strength of one anatomical connection was reduced to 20% of its normal value. The model comparison between normal and patient groups showed a significant group difference: the weakened anatomical connection in patients manifested itself as a weakened effective connection. This means that SEM has the potential for finding such connections in real brain fMRI data. But importantly, the feedback pathways which were downstream from the damaged anatomical connection for patients were also significantly reduced. This means that a weakened effective connection cannot necessarily be attributed to a weakened anatomical connection.? ? Remembering associations between names and objects is fundamental to language (see Horwitz & Wise 2008 for a review on using neuroimaging to study language). We examined the effect of retention interval and language on the neural correlates of memory for picture-sound associations and naming using fMRI and a paired associates (PA) task. Adults learned unique picture-sound associations. Ten animal photographs were filtered to be unintelligible but recognizable once shown the original image and were paired with sine wave speech versions of the depicted animal's name. We conducted fMRI while subjects, unaware of the embedded animals, performed a delayed PA task at two time points: immediately after training (D0) and 28 days after training (D28). Contrasting average delay period response, D0 was larger in posterior occipital and parietal cortices whereas D28 was larger bilaterally in temporal and frontal areas. To assess the language effect on PA memory retrieval, subjects were then trained on D28 to resolve the animals in the images and sounds and scanned performing the PA task in this informed state. Language use primarily reversed or increased lateralization of several foci in audio-visual association areas suggesting that naming relies on possibly separate but similar neural circuits to visual-auditory PA memory. These data will be used to expand our neural models so that they can perform the PA task (a review of the efforts at modeling memory processing is found in Horwitz & Smith 2008).
Showing the most recent 10 out of 32 publications
