Zvi N. Roth

Zvi N. Roth

using fMRI, eye-tracking, and modeling to uncover computational principles of vision

Publications

Representations in human primary visual cortex drift over time

Published in Nature Communications, 2023

The primary visual cortex is commonly believed to provide the rest of the brain with a stable representation of the external world. Here, we tested this assumption by analyzing a massive, longitudinal dataset of fMRI responses to naturalistic visual stimuli. In this dataset, subjects were scanned intensively over the course of an entire year, enabling a novel data analysis method inspired by computer vision and machine learning. This approach produced a comprehensive characterization of brain responses over the course of the year-long experiment. We found that visual representations exhibited continuous and cumulative change over many months. Our results challenge the classic view that the visual cortex is a static image-processing module, instead suggesting that even the earliest sensory areas of the brain are constantly changing and adapting over time.

Recommended citation: Roth, Z.N., & Merriam, E.P. (2023). "Representations in human primary visual cortex drift over time." Nature Communications. 14, 4422. https://www.nature.com/articles/s41467-022-34134-7

Natural scene sampling reveals reliable coarse-scale orientation tuning in human V1

Published in Nature Communications, 2022

How can we study orientation selectivity in humans? We showed that previous orientation studies were actually measuring stimulus vignetting rather than orientation tuning (Roth et al., eLife, 2018), but that doesn’t mean that there is no true orientation selectivity in the fMRI signal. It is possible that true orientation selective signals are present, but are eclipsed or overshadowed by stimulus vignetting. In this project I developed an approach to measure true orientation selectivity that is not vignetting. We used 2 image-computable models to analyze V1 responses in the Natural Scenes Dataset. One model accounted for stimulus vignetting, while the second model captured additional true orientation selectivity. Using this approach we managed to detect small but reliable orientation selective signals. We further revealed a coarse-scale radial map across visual cortex, a map that may underlie known perceptual biases.’

Recommended citation: Roth, Z.N., Kay, K., & Merriam, E.P. (2022). "Natural scene sampling reveals reliable coarse-scale orientation tuning in human V1." Nature Communications. 13, 6469. https://www.nature.com/articles/s41467-022-34134-7

Task-related activity in human visual cortex

Published in PLOS Biology, 2020

In a typical sensory neuroscience experiment, the neural response to an external visual stimulus is quantified, and an inference is made about the brain’s internal representation. Neural activity not related to the stimulus is considered ‘noise.’ However, unexplained variance also reflects a wide range of endogenous neural processes that have cognitive correlates. In a recent study aimed at uncovering cognitive processes underlying endogenous neural signals, I used multi-echo fMRI and ICA-based image denoising, physiological monitoring, pupillometry and eye-tracking, to characterize widespread activity across visual cortex. We uncovered activity that is independent of visual stimuli, yet time-locked to the ongoing task, and which we therefore termed ‘task-related’ activity. Similar global hemodynamic activity has been measured previously in monkeys using optical imaging, but the functional relevance of this signal has been hotly debated for a decade. I demonstrated that this task-related activity tracks arousal level, as indexed by both pupil size and heart rate. Computer simulations showed that arousal most likely decreases the temporal variability of task-related responses, consistent with a role in cognitive operations related to temporal attention.

Recommended citation: Roth, Z.N. Ryoo, M., & Merriam, E.P. (2020). "Task-related activity in human visual cortex." PLOS Biology. 18(11), e3000921. https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3000921&rev=2

Stimulus vignetting and orientation selectivity in human visual cortex

Published in eLife, 2018

The first research goal that I pursued during my postdoc was to develop and test a computational model that links measurements of neural activity in human visual cortex with perception. But unlike previous work, we included in the simulation the surrounding grey background, which provided a visual ‘context’ to the stimuli. Our model was not expected to have any orientation selectivity since the orientation channels from individual model neurons cancel out as a result of pooling. But when we included the visual context in the simulation, we found robust, large-scale orientation information in the model’s output. We argue that this orientation selectivity was entirely due to visual context, and we provide a mechanistic explanation for how it might manifest in the brain. This phenomenon, which we term ‘stimulus vignetting,’ is counter-intuitive because it implies that orientation selectivity reflects not only stimulus orientation, but stimulus contrast and shape as well. We used the model to predict neural responses to novel stimuli, and we confirmed these predictions with high-resolution, high-field strength fMRI.

Recommended citation: Roth, Z.N., Heeger, D.J., & Merriam, E.P. (2018). "Stimulus vignetting and orientation selectivity in human visual cortex." eLife. 7, e37241. https://elifesciences.org/articles/37241

Functional MRI Representational Similarity Analysis Reveals a Dissociation between Discriminative and Relative Location Information in the Human Visual System

Published in Frontiers in Integrative Neuroscience, 2016

Analyzing representations of relative location reveals that although early visual cortex represents stimuli in a highly location-dependent manner, the relative spatial representation (of one object with respect to another) there is distorted. On the other hand, representations in higher order regions preserve relative locations to a high degree. Computational modeling suggests that the unique representation in early visual cortex is a result receptive field properties, namely a mexican-hat profile, or negative surround. This demonstrates how the orderly neural organization across the cortex cortex results in a clear relationship between neuronal tuning, and the tuning of population measures such as fMRI voxels.

Recommended citation: Roth, Z.N. (2015). "Functional MRI Representational Similarity Analysis Reveals a Dissociation between Discriminative and Relative Location Information in the Human Visual System." Frontiers in integrative neuroscience. 10, 16. https://www.frontiersin.org/articles/10.3389/fnint.2016.00016/full

Position and identity information available in fMRI patterns of activity in human visual cortex

Published in Journal of Neuroscience, 2015

There is general agreement that visual processing is largely divided between a ventral and dorsal stream specializing in object recognition and vision for action, respectively. Here, we address the specific representation of viewed actions. Specifically, we study the degree of position invariance and hand/object manipulation specificity in the human visual pathways, characterizing the information available in patterns of fMRI activation during viewing of object-grasping videos, which appeared in different retinal locations. We found converging evidence for an information gradient within the dorsal stream of visual cortex: along the posterior–anterior axis, position information is gradually lost, whereas hand and action identity information is enhanced, leading to an abstract, position-invariant representation of viewed action in the anterior parietal cortex.

Recommended citation: Roth, Z.N., & Zohary, E. (2015). "Position and Identity Information Available in fMRI Patterns of Activity in Human Visual Cortex." Journal of Neuroscience. 35(33), 11559-11571. https://www.jneurosci.org/content/35/33/11559

Fingerprints of learned object recognition seen in the fMRI activation patterns of lateral occipital complex

Published in Cerebral Cortex, 2014

Using fMRI multivariate pattern analysis, coupled with an image degradation technique, I found that when degraded images became recognizable they evoked activation patterns in the lateral occipital complex (LOC), which were more similar to the patterns evoked by the intact (un-degraded) images. The evoked patterns also became less similar to the patterns evoked by pure noise image. These changes in brain activity occurred although the physical stimuli shown were identical before and after learning. These recognition-related effects were not observed in earlier visual areas. The results provide evidence that the coding in LOC reflects a perceptual-level representation of visual objects.

Recommended citation: Roth, Z.N., & Zohary, E., (2015). "Fingerprints of learned object recognition seen in the fMRI activation patterns of lateral occipital complex." Cerebral Cortex. 25(9), 2427-2439. https://academic.oup.com/cercor/article/25/9/2427/2926053