In nuclear medicine imaging, radiopharmaceuticals are used to study biological processes within the body. Radiation emitted from radiopharmaceutical decay is registered by a camera system, and reconstructed into images for interpretation of physicians. A central problem in nuclear medicine is achieving the highest image quality from the least amount of input data, since more input data often requires more radiopharmaceutical and thus a higher radiation dose to the examined patient. AI models for nuclear medicine image reconstruction have shown promising results regarding output image quality with significantly reduced input data. Both using AI generated synthetic data along with conventional image reconstruction algorithms, and by using completely AI based reconstruction algorithms. In this project, we are developing and applying AI based image reconstruction to nuclear medicine in two different contexts: 1. Radiation dose reduction for lung scintigraphy in pregnancy. Pregnant women are more likely than others so be investigated for suspected pulmonary embolism. Lung scintigraphy is one of the leading methods to rule in or rule out pulmonary embolism. By supplanting actual acquired input data with synthetic data, we hope to be able to significantly reduce the radiation dose to the patient and their fetus without compromising diagnostic image quality. To achieve this, we will train AI models on thousands of simulated lung scintigraphy examinations. 2. Tomographic image synthesis and biomarker development in neuroblastoma. Neuroblastoma is a type of childhood tumor with a very poor prognosis. The tumor cells overexpress noradrenaline receptors, allowing the use of mIBG scintigraphy for staging and treatment evaluation. Neuroblastoma treatment follows rigid protocols, to a large extent determined by semiquantitative scoring of mIBG scintigraphy examinations. The currently validated scoring systems (SIOPEN score, Curie score) only use two-dimensional data, even though most mIBG scintigraphy examinations are done with both whole-body 2D and partial-body 3D image acquisition. Current methods under-stage many patients, leading to insufficient treatment, and worse outcomes. We aim to develop a model that can create 3D whole-body images based in these input data, allowing more precise measurement of tumor extent and activity and development of better biomarkers.