A ROI-based algorithm for the classification of PET neuroimages

Positron Emission Tomography (PET) analysis aims to investigate the brain metabolism in order to associate brain functions with brain physical sub-areas activity. Studying PET images allows to measure actives biochemical processes in brain areas related to brain functions. Algorithms and software are available to analyze and study PET images: e.g., based on voxel-by-voxel (i.e., pixel areas in 3D images) analysis, as well as on regions of variation of the glucose in the brain (i.e., Regions Of Interest, ROIs). ROIs can be associated with specific pathologies. Nevertheless, (semi) automatic disease classification or support system tools using PET images, are still not available. In this paper we present a ROI-based analysis framework to support neurologists in PET images studies. The framework uses existing statistical tools to extract ROIs from PET images and a novel algorithm to compare ROIs of a blind pathological patient with a reference dataset. The algorithm has been implemented and tested on real cases.