Using independent components analysis to identify visually driven regions and networks in the human brain, using data collected during movie watching

Traditionally, regions involved in visual processing are mapped in the brain using simple localisers and/or anatomical techniques. As a more efficient (and interesting) alternative, Bartels & Zeki (2004) suggested that independent components analysis (ICA) could be used to segment the brain into functional regions, using data collected during movie watching. The first aim of this thesis was to explore the potential of this technique for reliable identification of visually driven regions and networks. In Chapter 2 I thoroughly and systematically explore the sensitivity of tensor ICA (TICA) to common pre-processing parameters and identify an optimal analysis pipeline. Despite some sensitivity of TICA to the parameters tested, robust components in visually responsive regions could be identified across outputs. Using an optimized pipeline, in Chapter 3 I demonstrate that visually driven components (in particular, peak voxels) are consistent across different samples and movie clips, supporting the use of this technique. In Chapter 4 I show that established resting state networks can be identified in an ICA analysis using movies, and that by increasing dimensionality sub-regions of these networks can be identified. Chapter 5 shows how these reliable components represented visual regions in the motion processing pathway. Based on the success of the technique at the group level, in Chapter 6 I apply the technique to individual observer data. Results show that functional networks and visual regions of interest can be reliably identified, supporting its use in future neuroscientific research. To address the short-comings of BOLD, the second aim of this thesis was to investigate whether MEG frequency data and fMRI bold data could be combined for analysis in a novel technique using TICA. First in Chapter 7 I address some prerequisites for a combined MEG frequency analysis using the technique. On the back of these results, I use the technique to generate interesting cross-frequency components (Chapter 8) and cross modality components using combined MEG and fMRI data (Chapter 9). These results show exciting promise for potential use in future neuroscientific work. In the final chapter, I investigate the potential use of ICA and changing dimensionality for mapping the functional hierarchy of the visual system. With development this could be a useful tool for understanding connectivity between sub-regions of functional networks. These results have important implications for the identification of visually responsive regions and for understanding neural activity during natural viewing