Non-invasive Measurement of Neural Inhibition

Almost all models of cortical visual processing implement some form of inhibition, such as lateral inhibition or response normalization. A problem in testing many such models is that we have no means of measuring inhibition in the human brain. In methods such as fMRI, inhibition may manifest itself as simply reduced signal amplitude. However, reduced fMRI amplitude reflects reduced excitation as well, so distinguishing between inhibition and reduced excitation is critical.

We believe the delay of the hemodynamic response function (HRF) may indicate the presence of suppression (Farivar, et al. 2011). The HRF delay is often neglected in fMRI analysis or treated as a nuisance variable. It is unlikely that the long delay that we reported—about 0.5s—is due to a neural delay. The lab is testing two potential causes of the delay: the delay may be due to (1) summation of two fMRI responses—a positive response from excited tissue and a negative response from inhibited tissue, or (2) neurochemical imbalance between excitatory and inhibitory neurotransmitters without separable fMRI components.

Our aim is to test these hypotheses using high-resolution fMRI to test the first possibility and MR spectroscopy methods to test the second possibility. Inhibitory interactions that characterize suppression will be induced either by visual or direct cortical stimulation. We will also assess the generality of the HRF delay in other perceptual suppression models and assess whether the HRF delay contains additional information regarding the underlying neural computations, regardless of its causes. 

Our working hypothesis of the hemodynamic delay and its relationship to local inhibition. We start with a statement that each voxel contains neurons that are competitively interacting. In the case of amblyopia, these would be neurons corresponding to ocular dominance columns representing the amblyopic and the dominant eye. When the dominant eye is closed, columns representing the amblyopic eye face very little suppression and, upon stimulation, can represent the signal and generate a positive fMRI response. However, when the dominant eye is opened but unstimulated, stimulation of the amblyopic eye forces the relevant columns to inhibit the columns of the dominant eye  in order to represent the stimuli. This would cause inhibition in the columns representing the dominant eye, and presumably result in a negative fMRI signal. The sum of these two signals--that observed from the voxel--would appear as an attenuated AND delayed response, similar to what we observed. 

Our working hypothesis of the hemodynamic delay and its relationship to local inhibition. We start with a statement that each voxel contains neurons that are competitively interacting. In the case of amblyopia, these would be neurons corresponding to ocular dominance columns representing the amblyopic and the dominant eye. When the dominant eye is closed, columns representing the amblyopic eye face very little suppression and, upon stimulation, can represent the signal and generate a positive fMRI response. However, when the dominant eye is opened but unstimulated, stimulation of the amblyopic eye forces the relevant columns to inhibit the columns of the dominant eye  in order to represent the stimuli. This would cause inhibition in the columns representing the dominant eye, and presumably result in a negative fMRI signal. The sum of these two signals--that observed from the voxel--would appear as an attenuated AND delayed response, similar to what we observed. 

REMOVING BARRIERS TO RECOVERY OF FUNCTION AFTER TRAUMATIC BRAIN INJURY

The most prevailing paradigm driving our rehabilitation efforts has been that factors local to the damage site regulate recovery of function—almost all current attempts at rehabilitation are focused on treating the injured tissue with, for example, tasks that target the lost function. The assumption is analogous to physical rehabilitation, where additional strength in a muscle is acquired by performing targeted physical exercises. This idea has been useful in driving recovery, but recovery is almost never complete, and the core of the idea that repairing the damaged tissue alone is necessary & sufficient for recovery is certainly incorrect.

No one knows if the rest of the brain integrates the recovered tissue. Normal activity in the brain is driven by complex and competitive interactions between multiple brain regions and neural modules. Once an area is damaged, this interaction is biased and altered and the recovering tissue (defined as a node in the cortical network) will receive substantially more inhibition than it can counteract. This additional suppression of weakened tissue prevents its recovery. We need information on how this interaction is realized, what neural properties are influenced by it, and how we can counteract excessive inhibition to reboot neuroplasticity for long-lasting recovery.     

Over the next three years, we aim to fill this knowledge void through a targeted, translational and trans-disciplinary approach that leverages a well-established model of neural function with a rigorous formalism of recovery to disentangle the two independent components of recovery—the node vs. the network. 

This schematic figure describes our paradigm. Below, a visual signal (a contour defined by aligned Gabor patterns) is presented to neurons in the primary visual cortex. Lateral interactions between the neurons empower efficient computation under normal conditions, but after injury, they are likely to become excessively inhibitory. In other words, healthy tissue is believed to excessively inhibit surviving weakened tissue, preventing its full recovery. Our studies aim to understand how lateral and top-down interactions are affected by injury and how re-balancing excitation and inhibition helps recover additional function. 

This schematic figure describes our paradigm. Below, a visual signal (a contour defined by aligned Gabor patterns) is presented to neurons in the primary visual cortex. Lateral interactions between the neurons empower efficient computation under normal conditions, but after injury, they are likely to become excessively inhibitory. In other words, healthy tissue is believed to excessively inhibit surviving weakened tissue, preventing its full recovery. Our studies aim to understand how lateral and top-down interactions are affected by injury and how re-balancing excitation and inhibition helps recover additional function. 

CORTICAL INTEGRATION FOR RECOGNITION OF OBJECTS FROM PURE DEPTH CUES

Late stages of the dorsal visual hierarchy are marked by an increase in neural selectivity for complex motion and binocular disparity signals, which may serve to represent 3D surfaces of objects. This is controversial, because object recognition is believed to arise strictly through hierarchical processing in the ventral visual pathway, effectively captured in computational models such as HMAX. Complicating the matter further is the fact that object-defining depth cues are each processed by different cortical mechanisms and are thus spread between the two parallel visual hierarchies—the dorsal visual pathway appears to be critical for disparity and structure-from-motion (SFM) processing, but the ventral visual pathway is important for shape-from-shading and texture.

Despite this division of labour, all depth cues appear to result in a common, depth-cue invariant representation. Depth-cue invariance allows the primate visual system to utilize multiple redundant sources of information while inferring object shape, a property especially important under compromised viewing conditions. Yet depth-cue invariance poses a problem for current models of object recognition—because depth cues such as SFM and disparity appear to be entirely processed by the dorsal stream, recognition of complex objects defined by these cues requires dorsal-ventral integration, in a manner that would by-pass the “classic” ventral object recognition hierarchy. 

In this project, we aim to better understand how depth information in natural scenes is encoded, how different depth cues are represented in the brain, and what the commonalities and differences are for object representations from different depth cues. 

 

Specifically, we are testing two models of cortico-cortical integration for object recognition from multiple depth cues. The integration model would predict that different depth cues, represented separately in different cortical areas, all send their signals to late stages of the ventral visual pathway for object recognition. The modularity hypothesis, on the other hand, predicts that all aspects of object recognition only make demands on the ventral visual pathway. 

The project is testing two competing hypotheses for object recognition from multiple depth cues. The integration model would predict that different depth cues, represented separately in different cortical areas, all send their signals to late stages of the ventral visual pathway for object recognition. The modularity hypothesis on the other hand predicts that all aspects of object recognition only make demands on the ventral visual pathway. 

The project is testing two competing hypotheses for object recognition from multiple depth cues. The integration model would predict that different depth cues, represented separately in different cortical areas, all send their signals to late stages of the ventral visual pathway for object recognition. The modularity hypothesis on the other hand predicts that all aspects of object recognition only make demands on the ventral visual pathway.