Background: In functional brain imaging, intersubject brain registration
is widely used to describe the loci of brain activation or lesions and to normalize
functional data between individual brains based on anatomical similarities. However,
such registration necessarily has limits because brain structure varies among
individuals and is not always closely correlated with brain function.
Objective: This study quantitatively compared three registration algorithms—linear
volume-based, nonlinear volume-based, and surface-based methods—using probability
and entropy maps of human visual areas.
Methods: fMRI retinotopic mapping was performed in 16 subjects to construct a
model for 12 visual areas. The surface and volumetric models of each visual area
were registered to the standard brain template using the three registration methods.
Results: After surface-based registration, the probability of visual areas being present in the common
space was increased approximately 3-fold compared with the volume-based method, but the average
probability was relatively small at approximately 0.3. On the other hand, average entropy was around
1 bit, revealing no significant difference between the two methods.
Conclusion: Our results indicate that the current technology has room for improvement and thus
should be used carefully with consideration of its limits. We suggest that the information-theoretic approach
can be naturally extended to the analysis of brain structure-function relationships by taking advantage
of mutual information.