ABSTRACT
Multiscale analysis of cerebral functional activation maps
Olivier Coulon
Brain mapping is a growing field of research which aims at determining
brain areas involved in the processing of cognitive or sensori-motor
tasks, and more recently at identifying interactions between these
areas. The number of projects has increased a lot in the last few years,
mainly because of the emergence of new modalities. These modalities
allow us to look in vivo at brain function: EEG, MEG, single photon
emission computed tomography (SPECT), positron emission tomography
(PET), and functional MRI (fMRI). Within these modalities, the last two
techniques (PET and fMRI) are the ones that are most used in brain
mapping projects.
Although the quality of PET and fMRI has increased, studies are still
limited by, first, the highly noisy nature of images, and secondly the
strong inter-subject anatomical and functional variability. These
problems have led to the use of analysis methods which are mainly based
on statistical processing using a large number of images. The nature of
these methods, making poor use of the spatial information, leads to a
loss of individual information for group analysis, and poor localization
power.
I will present here a brain activation map analysis method that I
developed during my PhD, which aims at processing group (multi-subject)
analysis while preserving individual information and overcoming
inter-subject registration limitations. This method is divided into 3
stages :
- computing of an object-based multiscale description of each individual
map: the scale-space primal sketch.
- building of a comparison graph matching all the primal sketches
computed from all the individual activation maps.
- group detection decision with a labelling process on the graph. The
label field on the graph is modelled by a Markov random field, and the
optimisation is therefore performed as the minimisation of a Gibbs
energy.
Result will be presented on a particular sensori-motor protocol.
Computer Science Home Page
Maintained by rbennett@cs.ucl.ac.uk