Quantification of 18F-FDG PET kinetic parameters using an image-derived input function and multimodal integration with resting-state fMRI metrics

Metabolic demand associated with resting-state brain activity is one of the main focus of neuroscience research. Task-free brain activation has been found to exhibit coherent spatial patterns, and the associated glucose consumption is predominant if compared to task activation. However, a complete characterization of the link between energy and function in the brain is still missing. The aim of this thesis project was to explore novel strategies for the integration between metabolic measures coming from Positron Emission Tomography based on fluorodeoxyglucose ([18F]FDG PET) and functional information extracted from resting-state Functional Magnetic Resonance Imaging (rsfMRI) measures. This was done adopting two different perspectives. On one hand, it was verified how metabolic and functional networks, inferred from time-series correlation across brain regions, relate to each other. On the other hand, across-subject similarity between sets of metabolic parameters and functional features was assessed. The analysis was performed on a dataset provided by Washington University in St.Louis, consisting of non-simultaneous PET and MR acquisitions on a large cohort of subjects. A first part of the work focused on [18F]FDG data. An Image-derived input function (IDIF) was extracted from the internal carotid arteries. This was later used for microparameter estimation with Variational Bayesian approach. Across-subjects correlation matrices were obtained for subjects series of K1 and k3 values. Moreover, average metabolic connectivity matrix was extracted from [18F]FDG parcel-level TACs. Similarly, from fMRI data, average functional connectivity matrix was extracted. Regional Homogeneity (ReHo) and Global Functional Connectivity (GFC) were estimated and across-subjects connectivity matrices were obtained for both parameters. Time-series connectivity matrices coming from both PET and fMRI images were used to assess similarity between metabolic and functional networks, whereas across-subject connectivity matrices were used to compare metabolic and functional parameters. To agevolate comparison, embedding was used on both timeseries and across-subjects connectivity: this was based on application of a gaussian kernel, followed by calculation of the Laplacian Eigenmaps, a nonlinear dimensionality reduction techinque. Resulting manifolds are called gradients in neuroscience, and are commonly used to study functional architecture in the brain. From a network perspective, metabolic and functional gradients exhibited significant correlation, and the regions in which they overlapped the most belong to visual and sensorimotor networks. Similar results were found between all combinations of [18F]FDG microparameters and fMRI features gradients, implying that both local and global functional relationship in the brain may be associated with specific metabolic fingerprints.