Protocol number 93-M-0170, NCT00001360 The Section on Functional Imaging Methods (SFIM) has advanced functional MRI (fMRI) methodology through development of processing and acquisition methods and research on the underlying mechanisms behind the fMRI signal. Our recent work has focussed on assessing ongoing cognition using ongoing fMRI connectivity profiles, assessing ongoing attention using either an assessment of ongoing connectivity or ongoing magnitude changes, further understanding and utilizing the gains in sensitivity obtained by massive trial averaging of single subjects, exploring the utility of multi-echo EPI for time series cleanup, and finally, imaging layer dependent activation and resting state fMRI fluctuations with high resolution fMRI. Due to limited space, we decided to highlight three of our projects that are representative of our ongoing work. Connectivity-Based Brain Reading Functional connectivity (FC) patterns in fMRI exhibit dynamic behavior on the scale of seconds with rich spatiotemporal structure and limited number of whole-brain, quasi-stable FC configurations or states recurring across time and subjects. Although several groups have hypothesized that these FC states relate to on-going cognition, and that their quantification may have clinical relevance, evidence of a direct relationship between FC states and on-going cognition has been missing. To fill this gap, we conducted experiments in which we scanned subjects continuously for 25 minutes as they performed and transitioned between 4 different mental states dictated by tasks (i.e., math, 2-back, visual attention, and rest) in blocks of 3 minutes. After appropriate pre-processing we computed FC states for windows ranging from 22.5s to 180s in length. We then submitted these FC states to a clustering algorithm and evaluated whether or not FC states grouped according to mental states. For subjects that performed the tasks consistently, the algorithm was able to group FC states according to mental states almost perfectly for all window durations. This was not the case for subjects with low and moderate performance. This first result suggests that short-term (<30sec) fluctuations in fMRI connectivity patterns can be reliably used to track ongoing cognition, on an individual subject basis, despite the noisy and indirect nature of BOLD signals as a marker of neuronal activity. We then conducted additional analyses to determine the spatial distribution of the most informative connections. We found that the distribution of the informative correlation changes were much more extensive that the task-related magnitude changes. This work was reported in PNAS. Classification of Distraction using Magnitude and Connectivity Changes We have started using the approach described above to develop an objective, continuous metric of sustained attention that is valid for a naturalistic task. We extract features of interest from changes in both fMRI magnitude as well as FC, and we build a machine learning classifier that uses these features to identify the presence of, and attention to, the auditory distractions in the task. Eight healthy adults each read text for 4-6 runs, with 30 9-line pages of text per run. On each page, either white noise or unrelated speech was played through headphones. Before each session, subjects were instructed to ignore the speech and focus on the reading (ignore speech trials) or attend to both (attend speech trials). Multi-Echo fMRI data were acquired during each session, and gaze data were collected using an infrared tracker. After pre-processing the data and applying multi-echo ICA denoising, we extracted features that could be used to distinguish white noise from speech trials or attend speech from ignore speech trials. The mean time-course in each region of interest (ROI) in a 200-ROI atlas was extracted, and Support Vector Decomposition (SVD) was used for dimensionality reduction. The remaining component time-courses became the magnitude Mag features used in classification. The correlations between ROI components in a 20-s sliding window were calculated and passed through dimensionality reduction to find the FC features. The features were sampled once per page and normalized. Logistic regression classifiers were built to classify (1) white noise trials vs. speech trials and (2) focus vs. split trials. In each case, classifiers were trained using (1) only Mag features, and (2) only FC features. When classifying white noise vs. speech, Mag features were effective and FC features were not. When classifying ignore speech vs. attend speech trials, however, FC features greatly outperformed Mag features, reaching an unusually high mean AUC value of 0.94. While we don't fully understand these results, this approach in general could pave the way for fMRI-based evaluations, interventions, or real-time feedback that could help readers especially those with attention-related disorders to better resist distraction and control attentional state. Layer Specific Mapping in Sensory and Motor Cortex at 7T using VASO The cortex consists of up to six cortical layers. Based on the different anatomically defined input-output characteristics across cortical layers, individual brain areas are expected to show different layer-dependent activity profiles according to their feed-forward/feed-back input. High-field high-resolution fMRI on a layer-dependent level might be able to elucidate afferent and efferent functional connectivity in healthy human volunteers using task activation or even resting state fMRI signal. Here, we report on an fMRI acquisition and analysis method that we are developing to measure layer-dependent activation and fluctuations. To account for limited signal-to-noise-ratio at sub-millimeter resolutions, acquisition and reconstruction strategies were optimized on a subject specific level, including custom designed RF-coil combination schemes, field of view geometry, application of FLASH-GRAPPA, and 7T field strength. We mapped layer-dependent activity using measures of cerebral blood volume (CBV) with VASO. We used the above approach for both a task (finger tapping vs finger movement without tapping) and for resting state fluctuation assessment. We found significant finger-tapping induced activity on a layer-dependent scale. The laminar distribution of task-induced activity and resting-state correlation results are observed to be specific to the unique input-characteristics for the corresponding tasks and resting-state seed regions. Functional MRI signal in upper layers of primary motor cortex (M1) have highest correlation values with resting-state fluctuations in primary sensory cortex, in line with the understanding that M1 receives its input from primary sensory cortex (S1) in upper cortical layers as opposed to the deeper layers. This result was further validated by comparing task-induced activity in M1 for finger-tapping involving finger tip touching compared to finger-movement without touch. The non-touching tasks differ in reduced exteroception in S1 and correspondingly reduced S1-input into M1. These results confirm the resting state results, namely that input from S1 into M1 is localized in upper cortical layers.
Showing the most recent 10 out of 84 publications
