Preview

PDF, English - main document Download (6MB) | Lizenz: Rights reserved - Free Access

Abstract

The identification of “neural biomarkers” of psychiatric disorders has been the aim of many recent fMRI studies. This endeavor, however, critically depends on the test-retest reliability of the corresponding fMRI task under investigation, which is an essential prerequisite for establishing associations between neural brain activation patterns and behavior. Previous research highlighted the low overall reliability of fMRI-based measures across studies and pointed towards the negative impact that might arise from computing difference scores between two correlated constituent task conditions, when applying fMRI tasks. This thesis investigated whether the test-retest reliability of the constituent conditions of an fMRI block design paradigm is higher than the reliability of the difference score, which is created by subtracting brain activation during one condition from another condition and is commonly used across a range of block design fMRI paradigms. All three empirical studies show that the reliability estimates of the constituent task conditions across three different block-design fMRI paradigms exceed the reliability estimates of the difference score. In addition, the presented empirical studies show that the level of correlation between the constituent conditions of the respective fMRI paradigm determined the level of the resulting reliability of the difference score. While moderate to good reliability of the difference score could be demonstrated in several brain regions and comparison of the test-retest reliability of the fMRI signal to the test-retest reliability of psychometric scales used to assess e.g. craving for food, in principle support the potential of fMRI of providing reliable measures, the presented results also challenge the application of difference scores when analyzing fMRI block-design tasks. This seems especially important in the light of the results of the first empirical study that demonstrated that the inter-correlation between the constituent task conditions of a difference score determine the magnitude of associations between the fMRI signal and external variables, and the precision with which effect size estimates can be derived from a specific fMRI study. A part of the limitations of fMRI-based measures that were identified by the presented empirical studies could be overcome in the future by establishing standards for fMRI data collection, data analysis and data interpretation, as well as using individual-level metrics to express brain activation, and by exploration and consideration of factors that influence test-retest reliability of fMRI block-design studies.