Kinetic Modeling Based Probabilistic Segmentation for Molecular Images

Abstract. We propose a semi-supervised, kinetic modeling based segmentation technique for molecular imaging applications. It is an iterative, self-learning algorithm based on uncertainty principles, designed to alleviate low signal-to-noise ratio (SNR) and partial volume effect (PVE) problems. Synthetic fluorodeoxyglucose (FDG) and simulated Raclopride dynamic positron emission tomography (dPET) brain images with excessive noise levels are used to validate our algorithm. We show, qualitatively and quantitatively, that our algorithm outperforms state-of-the-art techniques in identifying different functional regions and recovering the kinetic parameters.

• Type: Conference
• Year: 2008
• Where: MICCAI
• Authors: Ahmed Saad, Benjamin Smith 0002, Ghassan Hamarneh, Torsten Möller
• Keywords: Excessive Noise Levels, Medical Imaging, MICCAI 2008, Positron Emission Tomography, Self-learning Algorithm