Functional MRI Method Development

National Institute Of Mental Health

Investigators

Linked publications, trials & patents

Abstract

State-effects on resting-state fMRI functional connectivity Because of its simple setup (no need for stimuli presentation and/or recording devices) and its low demands in terms of subject compliance (keep still and awake), resting-state fMRI is often regarded as an ideal candidate for the development of fMRI-based clinical protocols. Unfortunately, excessively large levels of inter-subject and intra-subject variability in resting-state derivatives (e.g., FC estimates) have precluded this from happening. Part of this variability might be because current analytical methods assume that all resting-state subjects are in equivalent mental states, and disregard state-level effects (e.g., the contributions of the many diverse mental activities subject perform while resting) as sources of variability in resting-state fMRI. The purpose of this project is to evaluate this possibility and to quantify what level of inter-subject and intra-subject variability in static and time-varying functional connectivity estimates can be attributed to these state-level effects. For this purpose, we are currently working with a publicly available dataset that includes resting-state fMRI data annotated with descriptions of the content and form of thoughts that subject had while in the scanner. Preliminary results from this work show significant differences in static FC when subjects are grouped based on their scanning experience. Although such differences appear to be constrained to a small percentage of connections, the effect size is equivalent to that found between patient and healthy populations. Similarly, preliminary results suggest that some aspects of the content and form of subjects thoughts can be predicted based on their resting-state FC patterns. Overall, these results confirm that state-level effects are indeed an important contributor to resting-state fMRI variability and suggest the need for additional research to devise ways to model and account for them. Functionally Time-Resolved MRI Our section employs 7T fMRI, allowing us to develop novel methods that push spatial resolution into the sub-millimeter domain to resolve functional activity across cortical layers. One method being developed in our section is Functionally Time-Resolved MRI. This technique draws inspiration from innovations in the field of cardiac MRI for mapping concurrent physiological cycles (e.g. breathing and heartbeat) onto image dynamics by temporally resolving where in these cycles each data point has been collected. The aim of this development project has been to adapt these methods to brain imaging, where we can map the effects of similar physiological cycles, along with experimentally-induced brain activation patterns. Our efforts allow us to extract functional contrast from gradient-echo acquisition schemes that have been traditionally used for anatomical scanning. This means that we can push spatial resolution beyond the current state-of-the-art in fMRI while also acquiring multiple echoes. These methods have been presented at the international 2022 Organization for Human Brain Mapping (OHBM) meeting and are currently being written up as a manuscript. Representational similarities: angles or lengths? Representational Similarity Analysis (RSA) has become a standard tool in human and animal neuroscience. To understand the impact of choice of pattern similarity measure in RSA, we explored the similarity structure of brain patterns associated with a set of face images in two face-selective areas of the macaque brain (macaque data courtesy of Doris Tsao and Winrich Freiwald). The similarity structure observed with the cosine distance (an angular distance) matched the known form-of-tuning of neurons in both areas. In contrast, the Euclidean distance (sensitive to vector-length) revealed different similarity structures depending on the level of aggregation of the data. In other words, in the latter case RSA led to different conclusions about the neural representation of information depending on the degree to which neurons were pooled to form brain patterns. These results show the importance of distinguishing angles and lengths when interpreting pattern analyses, favor the use of angular distances, and helped re-evaluate an existing debate regarding homologies of face-selective areas in humans and macaques. This work was submitted as an abstract for SfN 2022. Orientation information in human primary visual cortex Information about the orientation of visually-presented gratings has been shown to be readily decodable from fMRI patterns measured from human primary visual cortex. A long-standing debate in visual neuroscience is whether the detected information reflects a property of the underlying neural populations or artifacts of the stimuli edge effects caused by interactions between the orientation of a grating and the border of the aperture used to display it. Predictions from computational models and pattern analyses of ultra-high field fMRI data favored the interpretation that orientation information detected with fMRI pattern analyses primarily reflect a property of the underlying neural populations, not the stimuli. Part of this work was presented at HBM 2022. Assessing and interpreting time-varying changes in fMRI signal This project, previously described in the 2021 Annual Report has now been completed. Results have been reported as a scientific publication in the journal NeuroImage titled: Ultra-slow fluctuations in the fourth ventricle as a marker of drowsiness Dimensionality Reduction of time series data During 2022 we have continued to work on this project previously described in the 2021 Annual Report. We are currently preparing a manuscript with a summary of the results which include estimates of intrinsic dimensionality for time-varying functional connectivity data (i.e., how many latent dimensions are needed to faithfully capture all structure present in the data) and guidelines for how to apply three state-of-the-art non-linear dimensionality reduction methodsnamely Laplacian eigenmaps, T-SNE and UMAPin a way that leads to behaviorally meaningful low dimensional representations of time-varying functional connectivity data. This work has been conducted in collaboration with members of the NIMHs Machine Learning Team. Application of Layer-Resolved fMRI at 7T Using the high spatial resolution in 7T, combined with cutting edge analysis approaches in layer fMRI, we are also interested to study the direction of information flow within the brain so-called feedforward and feedback signaling in conscious perceptual. Specifically, we are studying how the human neural mechanisms of visual perception and imagery (e.g., imagination or memory) may differ in layer fMRI signals that might suggest details on how the brain constructs perception of real-world objects and those of imagery. This work may have implications for understanding the cause of abnormal perception, including the hallucinations in psychiatric disorders. Multi-echo fMRI Our work on multi-echo fMRI processing methods has involved making existing methods more robust, as well as developing new ways to use multi-echo fMRI data. Our novel methods are focusing on better ways to combine external measures of noise from head motion, breathing, and heart rate patterns with the noise sources we identify using the information we gain from multi-echo acquisition. Preliminary results from this approach were presented at OHBM 2022. Our work on improving processing software was resulted in a peered reviewed publication that focuses on tedana.readthedocs.io. Additionally, as multi-echo fMRI has generated more interest, more extramural researchers are contacting SFIM staff for advice on how to design, acquire, and analysis multi-echo fMRI data for their studies.