Diapositivo 1

1
Temporal Correlation Compensation (Whitening) of
the fMRI data
R.J. Maximiano and J. Miguel Sanches Institute
for Systems and Robotics / Instituto Superior
Técnico Lisboa, Portugal

• Abstract
• BOLD-fMRI acquires the differences between
  oxygenated (HbO) and deoxygenated (HbO2) blood
  levels on the vessels of the brain.
• These differences are generally very small (5)
  relative to the signals magnitude and noised.
• Noise present in the fMRI data is temporally
  correlated (colored noise), making it harder to
  estimate active regions leading to more errors.
• A new method AR(M) was developed in order to
  remove the maximum possible of the
  autocorrelation present in these signals.
• In order to find the best model to decorrelate
  the signal, Akaike Information Criterion is
  integrated and sensibilities coefficients are
  continually re-estimated.
• Results (Monte Carlo tests) showed that AR(M)
  works well relative to other actual methods.

 
Problem Formulation
 
Real Data
This technique was applied on real data and
compared amongst AR(1), AR(M) and an image of
reference obtained by the software Brainvoyager.
Visual inspection suggests an increase of
precision and reduction of Type-1 Errors.
RecPad2010 - 16th edition of the Portuguese
Conference on Pattern Recognition, UTAD
University, Vila Real city, October 29th