Multimodal Naturalistic fMRI for Hypothesis-Driven Cognitive Neuroscience

Understanding how the brain integrates complex multisensory information in naturalistic contexts requires both ecologically valid paradigms and robust methodological tools. This thesis presents two complementary contributions to cognitive neuroscience: the development of a multimodal naturalistic fMRI database (NNDb-3T+) and the evaluation of a novel head motion reduction system (MR-MinMo). The NNDb-3T+ database was collected from 40 participants across two MRI sessions and one remote session. It combines full�length movie-watching, three sensory mapping tasks (somatotopy, retinotopy, tonotopy), synchronised eye-tracking, physiological recordings, and an extensive battery of behavioural and cognitive assessments. The database provides both raw and fully preprocessed data, enabling analyses of individual differences, multimodal integration, functional topographies, and naturalistic cognition. Technical validation demonstrates high data quality, including low head motion, accurate eye-tracking calibration, and reliable task-evoked responses, establishing NNDb-3T+ as a distinctive resource for reproducible research. The second contribution addresses one of the most persistent challenges in neuroimaging: head motion artefacts. The MR-MinMo device was designed to minimise head motion at the point of acquisition using a custom inflatable frame within the head coil. In a prospective study, MR-MinMo significantly reduced head displacement across multiple motion metrics during strong deliberate movements. A retrospective comparison of an independent database and NNDb�3T+ further confirmed its efficacy under long scanning sessions. While limitations remain, particularly in movements along specific axes, the device offers a promising hardware solution to complement software-based correction methods. Together, these contributions advance the field. The NNDb-3T+ enables hypothesis-driven research under naturalistic conditions, while MR-MinMo directly improves data quality