A fully-automated independent component classifier for reducing artifact in analysis of functional MRI

© 2015 Dr. Kaushik BhaganagarapuBrain imaging techniques, specifically, functional Magnetic Resonance Imaging (fMRI) has played a significant role in aiding our understanding of brain function. Taking advantage of the coupling between blood oxygen concentration and neural activity, fMRI has the ability to noninvasively map brain functions, providing researchers and clinicians new insights into the understanding of the human brain. However, interpretation of data obtained from fMRI experiments is challenging due to limited knowledge about the origins of the signal, low signal to noise ratio and the massive amount of data generated. Data driven techniques, specifically Independent Component Analysis (ICA) are increasingly being used to generate potentially valuable information on the nature of signal and noise in fMRI data. Unlike traditional methods for analysing fMRI data, where a design paradigm and assumptions about the haemodynamic processes are required, ICA offers a hypothesis free method to gain further insights in identifying the spatial location of brain activity. However, an enduring issue with data-driven analysis is the interpretation of results. To address this, we developed a novel algorithm, dubbed, Spatially Organised Component Klassifikator (SOCK), that identifies and removes noise from an ICA of fMRI data. The SOCK method does not require temporal information about an fMRI paradigm; does not require the user to train the algorithm; does not require querying external databases; requires only the fMRI images (additional acquisition of anatomical imaging not required); is able to identify a high proportion of noise related components without removing components that are likely to be of neuronal origin; can be applied to any fMRI studies; is automated, requiring minimal or no human intervention and is freely available in an open-source software package. In this thesis, we describe the SOCK algorithm and demonstrate its effectiveness in separating noise from signal by applying it to resting state, task-based and event related fMRI data, thus leading to additional insights in identifying spatial and temporal patterns of brain activity