Segmentation of dynamic PET images using cluster analysis

Abstract

Quantitative positron emission tomography (PET) studies provide in vivo measurements of dynamic physiological and biochemical processes in humans. A limitation of PET is an inability to provide precise anatomic localization due to relatively poor spatial resolution when compared to magnetic resonance (MR) imaging. Manual placement of regions-of-interest (ROIs) is commonly used in clinical and research settings in analysis of PET datasets. However, this approach is operator dependent and time-consuming. A semi- or fully-automated ROI delineation (or segmentation) method offers advantages by reducing operator error/subjectivity and thereby improving reproducibility. In this work, we describe an approach to automatically segment dynamic PET images using cluster analysis and we validate our approach with a simulated phantom study and asses its performance with real dynamic PET data. Our preliminary results suggest that cluster analysis can automatically segment tissues in dynamic PET studies and has the potential to replace manual ROI delineation for some applications.