• breath hold
• Cheyne-Stokes
• fMRI
• GLM

This work of thesis aims at studying neural correlates of respiration control with the goal of gaining knowledge towards the determination of the physiopathology of abnormal periodic breathing patterns such as Cheyne-Stokes respiration. Functional magnetic resonance imaging is used to investigate brain areas devoted to the closed-loop control system of respiration in healthy subjects, in particular those areas related to sensing of carbon dioxide levels and integration with peripheral chemoreceptors. Subjects were asked to perform about thirty-seconds-long voluntary apnoea with periods of about sixty seconds. The analysis is focused on the brainstem, thalamic nuclei and surrounding cortical areas, and originates from the work of Pattinson et al. (2009) to investigate the presence of nonlinear responses in carbon dioxide sensing networks, and their eventual role in allowing a separation between BOLD fluctuations caused by respiration-related neural activity and changes in the fMRI signal due to physiological noise from respiration or cardiac activity. A general linear model is applied to fMRI data, including static nonlinearities in the way described in Magri et al. (2011). The critical role of retrospective correction for physiological noise is deeply studied, in order to discuss on the optimal preprocessing path to be followed whenever in fMRI studies there is an interest over respiration-related brain activity. Methodological issues regarding correction for multiple tests were faced by testing both cluster-based methods and voxel-wise correcting techniques based on permutations. Finally, since the analysis produced a huge number of models, differing in terms of complexity and delays, different statistical model selection techniques are discussed, based on different comparison criteria such as adjusted R2 or BIC, testing also different approaches such as leave-one-subject-out cross validation, that may allow to reduce problems of overfitting. The study allowed to highlight activity in the thalamus, diencephalon, globus pallidum and caudate nucleus but failed to provide inter-subject consistent activations in the brainstem. The determination of some parameters defining region-specific responses, performed though model selection, may allow to determine time constants and the presence of eventual nonlinearities characterizing the mechanisms of the ventilatory drive. In this context, this work offers a framework that may inspire future works in approaching fMRI analysis with spatially-parametrized hierarchical linear models.