Methods for estimating metabolic brain connectivity at the region and voxel level using dynamic [18F]FDG Positron Emission Tomography

Over the last two decades, brain connectivity has become a dominant concept in neuroscience, and functional magnetic resonance imaging (fMRI) has significantly contributed to the understanding of the functional connectome of the human brain. However, fMRI only indirectly reflects neuronal activity through haemodynamic changes, whereas positron emission tomography (PET) captures relevant biological processes more directly, e.g., glucose metabolism. "Regions of the brain whose rCMRglc [cerebral glucose metabolic rates] values are significantly correlated are functionally associated, and the strength of the association is proportional to the magnitude of the correlation coefficient". This statement from a pioneering study by Horwitz et al. (1984) is the starting point for research on the so-called metabolic connectivity (MC), i.e., the set of relationships between the metabolic rates of different regions of the brain. MC is calculated on PET scans acquired with the glucose analogue [18F]fluorodeoxyglucose (FDG). However, instead of exploiting the temporal information of dynamic PET, most studies have used static measures to derive MC at the group level such as the covariation of metabolic information between subjects (subject-series matrices), usually with the standardized uptake value (SUV). This is in contrast with fMRI, where functional connectivity matrices are derived at the individual subject level from temporal correlations of brain region signals (time-series matrices). Therefore, a gold standard method to derive MC networks from dynamic [18F]FDG PET data at the single-subject level is missing in literature, and in this context the following thesis project is proposed. First, different approaches capable of retrieving MC matrices using dynamic [18F]FDG PET data were tested on a dataset of 71 healthy individuals provided by Washington University in Saint Louis, MO, USA. After pre-processing of the data (motion correction, coregistration to T1w image, parcellation with the Hammers anatomical atlas, filtering of the noisy initial frames, normalization with 5 different approaches to address the positive trend in the signal), different metrics were tested to calculate time-series MC: Euclidean similarity, Pearson correlation, Cosine similarity and Gaussian kernel. These were applied to both the non-normalized and normalized data. The same analyses were then performed on the concentration curve of the free tracer in the tissue and on the concentration curve of the tracer phosphorylated by the hexokinase enzyme, obtained following quantification using an image-derived input function and Sokoloff’s two-tissue compartment model. The obtained MC matrices were compared in terms of network structure and between-subject reproducibility. Subsequently, the time-series matrices were compared with subject-series matrices calculated as across-subject correlation of [18F]FDG parameters (i.e., SUVR, Ki, K1, k3) to assess the similarities between the results obtained through the proposed method and those derived from the standard approach. Further studies on the structure of the MC networks via graph theory and enrichment analysis with brain receptor maps were performed to better characterize the physiological underpinnings of the obtained MC networks. Finally, voxel-level MC analysis was tested using independent component analysis on [18F]FDG dynamic data (GIFT toolbox). The results demonstrated that, by using Euclidean similarity as a metric, normalisation of the TAC is not a necessary step. Furthermore, the MC matrices from the full TAC are strongly correlated with those of compartment 2, whereas compartment 1 is partially correlated with those of the first 20 min. Moreover, the graph metrics of the time-series matrices correlate positively with receptor/protein density information, whereas the metrics of the subject-series correlate negatively or are unrelated to those indices, which cast doubt on their physiological interpretation.