Modeling and Nonlinear Analysis in fMRI via Statistical Learning - Advance Image Processing in Magnetic Resonance Imaging - page 563

FIGURE 18.6 (See color insert following page 306 ). Effects on time course with varying

W-model for an activated pixel and a nonactivated pixel in the ST-SVR approach ( W-scale = 0).

different subjects (multisubject), or both. The additional data can increase the
sensitivity of the experiment and allow the generalization of any conclusion to
an entire population. A common technique for multisession analysis is to compute
activation maps for each session and then combine them into a composite through
ad hoc processing such as simple averaging. Limitations in this technique have
led to difficulties in the measurement of group differences, especially for more
subtle cognitive tasks. The ability of SVR to handle high-dimensional input data
makes it ideally suited for extensions to multirun and multisubject studies. The
ST-SVR formulation described in Subsection 18.4.1 and Subsection 18.4.2 allows
easy incorporation of data from multiple sessions by expanding the input vectors
and analyzing the data over multiple runs and multiple subjects together.
Similar to the spatial and temporal indices, now we have additional run and
subject indices, r and s . Suppose we would like to process fMRI data on S subjects
together and there are R runs for each subject, the input vector for data-driven
ST-SVR is then

xuvwt rrr rs s = [, , , , , , , ……… , , , , , , s ]

t

∈ℜ

4 ++ SR

(18.13)

12 i R 1 j S