Mosleh, Wissam2025-07-312025-07-312025-07-31http://hdl.handle.net/10393/50709https://doi.org/10.20381/ruor-31286Coronary artery disease (CAD) is a common type of heart disease, and leading cause of death worldwide. It can be reliably diagnosed and prognosed using myocardial perfusion imaging (MPI) by effective modelling of Myocardial Blood Flow (MBF) with Cardiac Positron Emission Tomography (PET) (1,2). The accurate quantification of MBF is made possible with accurate upstream image processing and in particular the localization and segmentation of the left ventricle. Moreover, accurate quantification of MBF is essential for diagnosis of coronary artery disease to guide optimal treatment. In this thesis, we develop an automated segmentation method for the left ventricle (LV) myocardium. Our research relies on 3D static volumes of relative myocardial perfusion images from the University of Ottawa Heart Institute (UOHI). We established ground truth manual segmentations using a semi-automatic process across multiple software. With these annotations, we were able to build and train a neural network that automatically outputs segmentations of the LV for perfusion PET images. Left ventricle myocardium segmentation can improve the reproducibility of PET MBF quantification, which will in return improve the diagnosis of CAD.enAttribution-NonCommercial-NoDerivatives 4.0 Internationalhttp://creativecommons.org/licenses/by-nc-nd/4.0/Convolutional Neural Networks (CNNs)82Rb PETMedical Image SegmentationMyocardial Blood Flow (MBF)Cardiac PETArtificial IntelligenceThesis