| |
Abstract:
Usually the model proposed for the Hemodynamic Response (HR)
in fMRI is limited to a predefined function such as Gaussian,
Poissonian or Gamma. This is not accurate enough to take into
account the variabilities among subjects and between areas in the
same subject, especially when the characteristics of the
Hemodynamic Response are to be used to reveal information on
physiological events accompanying neuronal activation. Here we
propose a general framework based on Support Vector Method (SVM).
allowing to approximate a function by using a small number of basis
functions chosen from a large set. The approximation scheme
fullfils two conditions, minimal approximation error and minimal
number of functions. The number of approximation functions could be
greater than one, thus capturing the multiphasic nature of some
Hemodynamic Responses seen in « blocked mode » fMRI
experiments. We applied SVM technique in experiments involving
patients with brain tumors located near rolandic sulcus and
submitted to motor task before and after removal of the tumor. The
results showed that up to three functions with different delays
were necessary to approximate the HR in activated areas. Moreover
delays were decreased after surgery suggesting cerebral
reorganization.
|
