Functional Imaging of Speech

In order to flesh out details of our modeling framework, our lab designs and carries out a variety of brain imaging studies in humans. Most of our work has concentrated on the use of functional magnetic resonance imaging (fMRI), a technique that is sensitive to changes in blood oxygenation levels that result from local brain activations. This method thus allows us to obtain statistical maps of cortical activations that can easily be compared with simulations from our modeling work. Because speech is only available for study in humans, fMRI is an invaluable technique, as it is very safe and completely non-invasive.

We are also using magnetoencephalography (MEG), another completely non-invasive experimental method, to augment and enhance our experimental data. MEG is a direct measure of brain activity in that it measures magnetic fields produced at the scalp by synchronous firing of pyramidal cells in cortex with millisecond resolution. One can also estimate the spatial locations of maximal activation in MEG, although the resolution of this method is much lower than with fMRI. We are currently working on tools to combine MEG results with fMRI results in the same subjects on the same behavioral task. Using regions of activation from fMRI as ‘seeds’ to help solve the MEG inverse problem, we will be able to look at fine time courses of activations in specific brain regions involved in speech.


Our lab has also developed a suite of region-of-interest (ROI) based analysis tools for fMRI that are available from the software page. We have shown that analyzing imaging results based on anatomically defined regions of interest across subject, rather than using spatial averaging, can significantly enhance statistical power.


The MGH10 dataset [linked here]:

10 subjects were scanned at the MGH/MIT/HMS Athinoula A. Martinos Center for Biomedical Imaging using a 3 T Siemens scanner and standard head coil. The data were inhomogeneity-corrected, afine-registered to the MNI152 template (Evans et al., 1992), and segmented using SPM2 software (Friston et al., 1995). The images were manually labeled by Tourville of Boston University using Ghosh’s ASAP software (Nieto-Castanon et al., 2003); the labeling protocol (Tourville and Guenther, 2003) is similar to the Cardviews protocol (Caviness et al., 1996), and has 74 labeled regions in this version, which is identical to the version used in the (Klein et al., 2009) evaluation study.

Subjects: 10 (4♀, 6♂)
Age: 22 – 29 (µ=25.3) years old
Volume: 256 × 256 × 128mm
Voxel: 1 × 1 × 1.33mm
TR: 6.6ms
TE: 2.9ms
flip angle: 8°

Imaging Facilities

fMRI studies are carried out at the NMR Center of the Massachusetts General Hospital.

MEG experiments are performed at the KIT/MIT MEG Joint Research Laboratory at the Massachusetts Institute of Technology.