Generalisable deep learning methods for low-dose positron emission tomography image processing

This thesis presents novel methods for integrating artificial intelligence into the medical imaging process to facilitate significant reductions in the radiation exposure caused from positron emission tomography: A gold-standard diagnostic nuclear medicine imaging technique widely used for diagnosis and staging in cancer and neurodegenerative diseases. Methods developed in this work address key issues surrounding artificial intelligence based image processing methods related to performance on previously unseen data and accurately recovering image characteristics essential for accurate diagnosis.