Fundamentals of Data Analysis Methods in Functional MRI - Advance Image Processing in Magnetic Resonance Imaging - page 484

15.5 DATA-DRIVEN ANALYSIS OF FUNCTIONAL

TIME SERIES

Data analysis with the aforementioned voxel-based methods is limited to the
detection of cortical activity with a strictly task-related temporal behavior. With
other processing strategies, such as principal component analysis (PCA) [40] and
independent component analysis (ICA) [41, 42], information contained in time
courses is extracted without strong a priori hypotheses about the time profile or

the spatial extension of the cortical areas (see Chapter 18 ). These methods are

based on the decomposition of the intrinsic spatiotemporal structure of the fMRI
time series in orthogonal spatial patterns or eigenimages (PCA), and in indepen-
dent component maps (ICA) with different time courses. The drawback of PCA
and ICA consists in the difficulty of giving a physiological interpretation to the
great number of different components [41, 42].

15.6 COMBINING BRAIN FUNCTION

AND ANATOMY

15.6.1 C OREGISTRATION OF F UNCTIONAL AND A NATOMICAL

D ATA S ETS

As mentioned in the introduction, a typical fMRI measurement session includes,
before or after the collection of the functional time series, the collection of
anatomical images covering a region of interest or the whole brain. These images
are used for the spatial normalization of the data in a standard space and also for
a better visualization of the statistical maps. Anatomical images may be collected
using 2-D T1-weighted sequences (2-D anatomical reference) with the same
spatial parameters (position, field of view, thickness) as the functional volumes.
In this case, assuming that there is no significant subject motion in the interval
between the functional and anatomical acquisition, coregistration between func-
tional and anatomical volumes can be obtained simply by superimposing, for
each volume of the time series, the stack of functional slices on the stack of
coplanar anatomical slices. Functional maps can thus be overlaid either onto the
anatomical stack of slices or functional stack of slices.
However, if more sophisticated visualizations of functional maps are to be
obtained (e.g., using folded or morphed reconstructions of the subject's cortex; see

Subsection 15.6.3), anatomical reference images of the subject's whole brain are

typically collected using 3-D (e.g., with three encoding gradients) T1-weighted
sequences. These sequences provide very good spatial resolution (1 mm ×

1 mm) and high contrast between gray and white matter in a relatively short acqui-
sition time (8-20 min for whole brain imaging). When these 3-D anatomical images
are collected, coregistration of functional and anatomical data sets is obtained using
information on the MR-scanner slice position parameters of the T2*-weighted mea-
surements (number of slices, slice thickness, interslice gap, in-plane resolution, field
of view, angles and offsets in the readout, phase and z directions) and on analogous

× 1 mm