ANALYSIS OF INDIVIDUAL BRAIN ACTIVATION MAPS USING HIERARCHICAL DESCRIPTION AND MULTISCALE DETECTION

We propose a new method for the analysis of brain activation images that aims at detecting activated volumes rather than pixels. The method is based on Poisson process modeling, hierarchical description, and multiscate detection (MSD). Its performances have been assessed using both Monte Carlo simulated images and experimental PET brain activation data. As compared to other methods, the MSD approach shows enhanced sensitivity with a controlled overall type I error, and has the ability to provide an estimate of the spatial limits of the detected signals. It is applicable to any kind of difference image for which the spatial autocorrelation function can be approximated by a stationary Gaussian function.