Methodological Aspects of Clinical fMRI - Reliability Assessment and Development of Data Analysis Strategies

Sammanfattning: Functional magnetic resonance imaging (fMRI) is a non-invasive technique used for the investigation of brain function, which has found numerous applications within basic neuroscience. The introduction of fMRI as a clinical tool for presurgical mapping of pertinent cortical regions in patients with tumour or epilepsy has facilitated neurosurgical planning, and most likely reduced the risk of severe postoperative deficits. However, there is a fundamental difference between the application of fMRI in experimental research and in the clinical setting. While inference is often drawn from data on group level in neuroscientific applications of fMRI, the clinical use of the method demands that reliable results be obtained in individual patients. The aims of the methodological studies presented in this thesis were to increase the reliability and extend the usefulness of clinical fMRI. In one study, gradient-echo field maps were utilized to assess the sensitivity of echo-planar imaging to the blood oxygenation level dependent (BOLD) contrast. Theoretical expressions for the calculation of the BOLD sensitivity were verified and improved. The BOLD sensitivity was investigated in a group of healthy volunteers using a clinical magnetic resonance imaging (MRI) system, and it was concluded that the field map method accurately predicts BOLD sensitivity. In another study, a flexible model was introduced in order to increase the confidence of clinical fMRI examinations of patients unable to fully comply with a typical clinical fMRI experiment. The method was applied to experimental and simulated data, and then used to retrospectively analyse patient data. The conclusion drawn from this study was that the proposed flexible model improves the detection of activation in partially non-compliant subjects. In the third study included in this thesis, the model-free algorithm locally linear embedding was applied to fMRI data analysis. The proposed data-driven algorithm was optimized and investigated with respect to reliability and possible benefits in a clinical setting. The algorithm was found to compare well to the traditionally used method of principal component analysis, and showed benefits when applied to simulated fMRI data exhibiting non-linear characteristics. Finally, the test-retest reliability of resting-state fMRI was compared with the reliability of traditional task-based fMRI. Resting-state fMRI compared well with task-based fMRI experiments, thus possibly extending the use of fMRI to patient groups that have hitherto not been able to benefit from fMRI examinations.