Skip Navigation

Attenuation of Task-Correlated Motion Artifacts During Speech

Author: Gopinath Kaundinya, PhD

A number of fMRI paradigms require overt word generation in the MRI scanner. Speech involves local nonrigid movement of pharynx, tongue, and jaw, which have been demonstrated to change the magnetic field distribution within the brain. The greatest magnetic field changes due to speaking occur in the inferior, temporal, and frontal regions of the brain, decreasing rapidly superiorly and posteriorly. Voxels in the inferior, temporal, and frontal regions may exhibit more than 100% task-correlated signal changes. Since these artifacts are temporally locked to task-performance they show up strong false positive activation in fMRI statistical parametric maps. Utilizing the differential temporal profile (see Figure 2) of task-correlated motion (TCM) artifacts and blood oxygenation level dependent (BOLD) MR signal changes we constructed a classification algorithm that was able to selectively detrend TCM artifacts while keeping BOLD activation intact in event-related speech fMRI paradigms. Figure 1 depicts the results of application of the selective detrending method on a left lateral sagittal slice. Voxels with blue overlay in the left figure are significantly activating TCM artifacts. These artifacts are removed after application of the selective detrending (right) algorithm developed in our lab which leaves BOLD activation (red overlay) in inferior frontal gyrus and precentral and motor areas intact.

We are currently working on extending this method to block word-generation paradigms, which will immensely benefit more complex speech fMRI studies.

Figure 1

Figure 1: (left) FMRI statistical parametric map showing task-correlated motion (TCM) artifacts (blue) and brain activation in red during a semantic word-generation task. (right) The selective detrending method removes TCM artifacts and leaves brain activation intact.

Figure 2

Figure 2: Estimation of TCM IRFs from a lateral sagittal slice (top) and medial sagittal slice (bottom) for a speech fMRI dataset. Activation map, thresholded at GLM R2 > 0.16, is overlaid on low-resolution spiral FMR images. The green arrows point from clusters of voxels surmised to be TCM-corrupted and screen-grabs of the corresponding 2 X 2 voxel grid TCM impulse response function (IRF) graphs. The blue arrows are drawn between putative BOLD voxels and their corresponding BOLD IRFs. The horizontal axis for each of the individual voxel plots runs from 0 to 24.9 s. The vertical scales for all the individual plots are marked in units of fractional signal change.

Project Team

Kaundinya Gopinath (PI), TBA Postdoc