Characterization and compensation of systematic noise in functional magnetic resonance imaging.

摘要

Functional magnetic resonance imaging (fMRI) has emerged as an important tool for noninvasive neuroscientific research. A limit to its effectiveness, however, is the presence of systematic noise that can obscure neuronal activation.; Systematic noise in fMRI has a temporal and/or spatial structure, as opposed to additive random Gaussian white noise (e.g. thermal fluctuations). Several examples are low frequency signal drifts, head motion, physiological noise, and spontaneous neuronal events. These systematic noise sources are generally multiplicative and depend on the signal strength. As the fMRI signal is increased, by increasing voxel size or field strength, these noise sources may dominate the thermal noise, and determine the effective signal-to-noise ratio of a functional imaging experiment. Thus, understanding these noise sources and how to mitigate their effects is an important step in maximizing the potential of functional MRI as a neuro-imaging tool.; This dissertation investigates characterization and compensation techniques for several types of systematic noise in fMRI. First, mitigation techniques for signal drift in single cycle MRI studies and physiological noise (caused by the respiratory and cardiac rhythms) are investigated, with functional contrast increased using appropriate noise compensation. Then, the effect of physiological noise in multi-shot imaging is explored. It is seen that the effective repetition time (TR) combines with the frequency of the physiological noise to modulate the level of physiological noise variance induced in a multi-shot study. A noise compensation process is next applied to a rapid, multi-slice acquisition and is shown to reduce noise variance down to the level of the associated single-slice case. Finally, resting state low frequency functional connectivity patterns are examined. Using a multi-echo sequence, they are shown to have the same T₂* and echo time dependence as “normal” task activation. A data-driven method of detecting functional connectivity patterns using a clustering algorithm is also investigated, and compared to the standard reference-based approach.