The extended axial field-of-view (FOV) of so-called total-body positron emission tomography (PET) scanners provides an opportunity to image the dynamic radiopharmaceutical distribution across multiple organs at the same time. Parametric maps can be derived from 4D TB-PET data through kinetic modelling analyses, and these images can offer clinicians additional insight on how to guide and manage treatment. However, the implementation of dynamic TB-PET imaging protocols into clinical routine is still limited because of their long acquisition time. Additionally, time-activity curves have high noise levels which inevitably results in poor quality of parametric images. This project aims to improve 4D TB-PET imaging performance by applying deep learning (DL) techniques for image enhancement. The proposed project will look into DL-based methods for denoising of 4D TB-PET datasets in the spatial (image) and temporal domain, with the goal to improve data quantitation of dynamic TB-PET studies. The goal is to have an intelligent and time-efficient (DL-based) model to optimize 4D TB-PET protocols (both in clinical and preclinical settings) by generating high quality parametric images directly from early acquired dynamic time frames.