Manual preprocessing and analyzing of functional magnetic resonance imaging (fMRI) datasets can be a cumbersome endeavor. Using readily available processing pipelines can help with this, but such an approach bears many risks.
This thesis first describes the nature of event related fMRI datasets, what it measures and how such data can be preprocessed and analyzed. After listing the main standard neuroimaging toolboxes used to process event-related fMRI datasets, this thesis describes the multiple issues connected with the current processing approaches and the general challenges the field tries to tackle to innovation in technology and methodology.
The four issues identified in this thesis are: (1) inaccessibility and stickiness of neuroimaging toolboxes, (2) missing general standards for neuroimaging analyses, (3) a reproducibility and transparency crisis and (4) insufficient data quality control and results reporting. The three challenges identified in this thesis are: Due to innovation (1) in the spatial dimension, (2) in the temporal dimension and (3) in signal processing.
The first study (Study A) tackles most of these issues and challenges by introducing a new neuroimaging toolbox, called fMRIflows. This toolbox is a consortium of fully automatic processing pipeline capable of performing state-of-the-art data preprocessing, as well as first- and second level univariate and multivariate analysis. Validation of the toolbox is done by analyzing three different fMRI datasets with different temporal resolution (i.e. 2000ms, 1000ms and 600ms) and comparing the output created with fMRIflows to the output created with state-of-the-art neuroimaging software packages fMRIPrep, FSL and SPM. The validation shows that no strong difference between the output of the four toolboxes can be observed. Furthermore, the study shows that an adequate temporal filtering of an fMRI dataset with a sub-second temporal resolution can lead to improved temporal signal-to-noise-ratio (TSNR) after preprocessing and an increased statistical sensitivity in the 1st and 2nd level analysis.
The second study (Study B) tackles many of the beforementioned issues, but focuses on the 4th (i.e. results reporting) in particular, by introducing a new neuroimaging toolbox, called AtlasReader. This toolbox can be used to generate coordinate tables, region labels and informative figures from statistical MRI images. Study B successfully introduces a neuroimaging toolbox that allows to create beautiful and informative results reports, independent on the operating system of the user. Furthermore, AtlasReader allows the extraction of association tables from multiple atlases which usually are not accessible to a single operating system.
The third study (Study C), uses these the toolboxes developed in Study A and Study B and show their application in a cognitive neuroscience study in the domain of multisensory integration. Study investigated the brain mechanisms involved during the encoding and subsequent retrieval of semantically congruent multisensory objects. In this study we found that the low-level visual cortex reliably can decode whether an incoming visual stimulus previously had been perceived in a semantically congruent or incongruent context, even if the visual stimuli was only perceived once before. The results from this study further support the notion that the low-level visual cortex has multisensory architecture and that the creation of memories profits from a multisensory semantic congruent stimuli exposure.
As a next step, the thesis assesses the first two studies with respect to the previously mentioned issues and challenges and the third with regards of novelty and new insights gained from this cognitive study. Followed by a critical assessment with regards to the studies limitations. After that, the future directions of fMRI processing analysis routines and of fMRI-based investigation of multisensory integration is discussed. The thesis is finished with a prospect section of what might come and the general conclusion.
Together, the studies comprised in this thesis highlight and address the issues and challenges currently present in the neuroimaging domain and provide a path forward. Furthermore, the findings and outcomes of all three studies contribute to a better understanding of how to correctly preprocess and analyze fMRI datasets, as well as how the mechanism behind multisensory integration takes place.