Local Means Analysis: an unsupervised method for the segmentation of rodent whole-body dynamic PET images

A new method for segmentation of PET images is presented, dedicated to rodent whole-body images. The method is based on an analysis of the time activity curves in the core of the organs, where they are the least affected by spillover and physiological movements. The method is validated on 144 simulated PET sequences. Whatever the number of organs and the organs´ pharmacokinetics, the method achieves a good segmentation quality and a remarkable detection rate, even under condition of tough spillover. On biological study on rats, the method correlates well to manual delineation, while allowing the delineation of a higher number of organs than manual delineation together with a better confidence on the measure