Abstract

Coronary heart disease is the most common cause of death in the world and non-invasive medical imaging techniques like positron emission tomography (PET) are important to help diagnose a patient suffering from atherosclerosis as early as possible. Unfortunately, dynamic PET measurements using radioactive water to examine blood flow create challenging image reconstruction and parameter estimation problems.Non-negative matrix factorization (NMF) has been successfully used as a data analysis tool for many different applications. In this thesis, we will motivate the use of NMF through model-based approaches to dynamic PET reconstruction and examine the results and performance of different NMF algorithms when applied to dynamic PET measurements with very poor statistics.