Vector-based active surfaces for segmentation of dynamic PET images

Dynamic PET imaging enables the study of radiotracers concentration kinetics along time. However, PET images suffer from limited spatial resolution, low signal-to-noise ratio, and time dependent contrast between tissues, making segmentation of biological volumes difficult. Deformable models are of great interest due to their inherent boundary continuity, but their applications to PET images still remains challenging due to high sensitivity to noise. To address these limitations, we propose a method to perform 3-D+time dynamic PET image segmentation using parametric active surfaces based on a gradient of the vectorial image. This method takes advantage of both spatial and temporal consistency of the signal along the acquisition. We validate our method using GATE Monte Carlo simulations and compare it with single frame based active surface models. We show significant improvement of several figures of merit.