Multiple Scales of Stimulus Representation in the Human Brain

National Institute Of Mental Health

Investigators

Linked publications & trials

Abstract

In FY23, we continued to pursue two lines of research, focusing on 1) characterizing the representation of stimulus features in primary visual cortex and 2) studying how these stimulus representations interact with feedback signals to visual cortex associated with general arousal, emotional valence, and object-selective processes. Studies were carried out under clinical protocol NCT00001360. 1) Characterizing stimulus representations in human visual cortex. A central goal in the laboratory is to understand the neural computations that give rise to stimulus feature selectivity. Many of the studies in our group use the visual system as a model, and specifically study the computations that lead to selectivity of for visual stimulus features, such as orientation, spatial frequency, and direction of motion. Our research has focuses on these features because they are hypothesized to depend on a set of core 'canonical' computations that are carried out throughout the brain. Understanding these canonical computations in visual cortex, we hypothesize, will provide important insights into a broader set of brain functions in both health and disease states. Stimulus orientation is one of the most basic stimulus features represented in primary viual cortex (V1). Yet, after more than 50 years of research, the representation of orientation is inadequately understood. It is well established that orientation-selective V1 neurons are organized at a fine spatial scale in a pseudo-periodic, columnar structure across the cortical surface. However, using functional magnetic resonance imaging (fMRI), we have previously discovered an additional level of organization, a coarse-scale orientation bias in which each fMRI voxel in V1 exhibits an orientation preference that depends on the region of space that it represents. We have recently developed a population model of V1 that provides a computational framework for investigating orientation selectivity (Gardner and Merriam, 2021). Using the model, we have demonstrated that three theoretically distinct neural mechanisms could each contribute independently to orientation biases: stimulus vignetting, neural gain fields, and asymmetric surround suppression. This modeling result shows that these mechanisms can easily be confused with one another. For example, attempts to measure surround suppression, which is thought to be a marker for excitatory/inhibitory imbalance in cortex, and implicated in a wide range of neuropsychiatric disorders, could in fact be confounded with other neural mechanisms. One important implication of our model is that it should be possible to isolate orientation-selective signals arising from each of these factors independently. Importantly, our modeling work suggests that the relatively low signal-to-noise ratio of fMRI may limit the ability to achieve this. To address this, we fit our model using data from a massive, longitudinal dataset of fMRI responses to 30,000 naturalistic scene stimuli (the NSD dataset). This massive fMRI dataset, collected with a 7 tesla high-field strength scanner with many repeated measurements from a small group of participants (> 30 sessions from 8 subjects), has provided new insights into the relative contribution of each of these mechanisms for orientation selectivity (Roth et al., 2022). Next, we analyzed these data longitudinally, testing whether neural computations changed over time. We discovered progressive, cumulative drift in neural representation in visual cortex (Roth & Merriam, 2023). Our work on understanding neural computations in the intact visual system is vital in determining how computations are altered in psychiatric and neurological disorders. 2) The role of feedback in visual processing A second theme in the laboratory is to understand the influence of non-sensory cognitive signals in visual perception. We have identified a type of brain activity that reflects a subjects general engagement in a task. This task-related activity contributes prominently to brain hemodynamic responses measured with fMRI, is independent of external visual stimuli, and instead reflects internal brain states. Task-related activity appears to be distinct from spatial attention in that it is global in cortex rather than retinotopically-specific. This activity may be related to general arousal. To test this hypothesis, we have developed a task that enabled us to isolate task-related activity from visually-evoked activity, and to systematically vary 1) task difficulty, 2) the expected reward for correct performance, and 3) the temporal predictability of the task structure. We found that a widespread fMRI response tracks arousal level in each of these task conditions, exhibiting a significance correlation with pupil size and other physiological correlates of arousal (e.g., heart rate variability). We have recently shown that measurements of pupil size can be decomposed into separate attention- and arousal-related components and that the arousal component closely tracks the task-related fMRI response. We have published two studies on this work (Roth et al., 2020; Burlingham et al., 2022), and we are currently have a third manuscript under review. Understanding the neural mechanisms of global brain states has implications for the study of a number of neurological and psychiatric disorders, including schizophrenia, autism and Attention Deficit Hyperactivity Disorder (ADHD).