Methods for non-invasive [18F]FDG PET multiparametric imaging in brain and total-body scans

Positron Emission Tomography (PET) with the 18-fluorodeoxyglucose ([18F]FDG) radiotracer enables in vivo quantification of glucose metabolism in tissues. The most informative approach to quantify PET scans is compartmental modeling, which requires dynamic scans. When applied at the voxel level, this approach produces highly detailed maps of tracer kinetic parameters, resulting in multiparametric imaging. Yet, in clinical practice and most research settings, [18F]FDG PET is typically performed as a static scan, analyzed using the Standardized Uptake Value (SUV), which is easier to obtain. For [18F]FDG in the brain, SUV has been shown to provide an adequate approximation of K_i (ml/cm3/min), which represents the net [18F]FDG uptake rate. However, SUV is susceptible to many sources of non-biological variability. Moreover, it cannot provide specific information like kinetic parameters K₁ (ml/cm3/min), which reflects tracer transport from blood into cells, and k₃ (min-1), which represents [18F]FDG phosphorylation by hexokinase. One of the main factors limiting the application of PET multiparametric imaging is the need to assess the arterial input function (AIF), generally obtained through invasive blood sampling. One alternative option is the Image-Derived Input Function (IDIF), which estimates the AIF from vessels included in the image field of view (FOV). This approach is relatively straightforward in scans with large blood vessels, such as cardiac scans, but is more challenging in brain scans, where the only feasible extraction sites are the carotid arteries, which are small and subject to partial volume effects. Recently, Long Axial Field of View (LAFOV) scanners have been introduced. These scanners feature axial FOVs ranging from 1 to nearly 2 meters, enabling dynamic total-body acquisitions. As a result, large blood vessels, such as the aorta, can be used for compartmental modeling in almost any organ, facilitating non-invasive multiparametric imaging. However, due to the high degree of physiological heterogeneity across tissues, conventional compartmental modeling approaches, typically based on applying a single model within a predefined volume of interest, are often inadequate to fully exploit the potential of these new technologies. The general aim of this thesis is to facilitate non-invasive [18F]FDG multiparametric imaging, addressing the limitations of conventional brain scans and the methodological gaps associated with total-body scans. To enable non-invasive [18F]FDG multiparametric imaging in brain scans, this thesis presents an open-source toolbox that performs automatic IDIF extraction from dynamic [18F]FDG brain PET images. The toolbox is adaptable to different scanner characteristics, as it does not rely on any specific technology or multimodal imaging approach for vessel identification. Additionally, it implements a blood-free method for partial-volume correction of the IDIF. Furthermore, a quantification framework based on this toolbox is applied to describe different metabolic trajectories associated with the early stages of Alzheimer Disease. Concerning total-body PET, several data-driven approaches for [18F]FDG multiparametric imaging, which do not require definition of a specific model structure, are applied to scans coming from the uEXPLORER LAFOV scanner. These methods include Spectral Analysis (SA) and graphical approaches, consisting of Patlak analysis for K_i estimation, applied to late-phase kinetics, and a linearized model for K₁ estimation, applied to early-phase kinetics. Among the tested methods, SA is the most flexible, but it is prone to overfitting. To overcome this limitation, a novel Bayesian regularization approach for SA is presented and preliminarily tested on both simulated and real data. Overall, this thesis may help extend the applications of [18F]FDG multiparametric imaging in research and support the translation of these approaches into clinical practice.