Master of Science
Menon, Ravi S.
Electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) are complementary modalities commonly acquired simultaneously to study brain function with high spatial and temporal resolution. The time-varying gradient fields from fMRI induce massive-amplitude artifacts (GRAs) that overlap in time and frequency with EEG, making GRA removal a challenge for which no satisfactory solution yet exists. We present a new GRA removal method termed Schrödinger filtering (SF). SF is based on semi-classical signal analysis in which a signal is decomposed into a series of energy-based components using the discrete spectrum of the Schrödinger operator. Using a publicly available dataset, we compared our pipeline, which features only the popular average artifact subtraction (AAS) technique and SF, to two popular pipelines. The SF pipeline outperformed across all frequency bands based on metrics of signal preservation and GRA removal. SF, when combined with AAS, is therefore superior for removing GRA from EEG data.
Summary for Lay Audience
Electroencephalography (EEG) directly measures brain activity with electrodes placed on the scalp. EEG records measurements very quickly although it is unable to well-localize the sources of the activity. Functional magnetic resonance imaging (fMRI) forms a set of images of the brain over time. These images measure changes in blood flow and oxygenation that accompany brain activity. Therefore, fMRI indirectly measures brain activity. fMRI well-localizes brain activity but takes relatively long to acquire a single image. Both EEG and fMRI are non-invasive. The combined modality of simultaneous EEG and fMRI (EEG-fMRI) therefore offers the benefit of noninvasively recording brain activity with both high spatial and temporal resolution.
One unique challenge of EEG-fMRI is the gradient artifact: a large-amplitude set of signal disruptions in the EEG data caused by the interaction of the fMRI magnetic fields with the EEG equipment. The gradient artifact has been studied for over a decade and numerous solutions have been proposed. However, no solution reduces the gradient artifact while preserving the underlying signal such that it is not a significant impediment to the analysis of the EEG data.
We present a new technique for removal of the gradient artifact called Schrödinger filtering. Schrödinger filtering is able to decompose a signal into a set of constituent signals, each possessing a different energy, where energy is proportional to signal amplitude. Schrödinger filtering is well-suited for gradient artifact removal because the gradient artifact is greater in energy than the EEG signal.
On an online dataset, we applied a popular gradient artifact removal step called average artifact subtraction followed by Schrödinger filtering. We compared the performance of our processing pipeline to that of two other popular pipelines in terms of signal preservation and artifact removal. Our pipeline outperformed the other two. These results indicate that Schrödinger filtering is a superior processing technique for gradient artifact removal that helps with the analysis of EEG data of EEG-fMRI.
Benigno, Gabriel Bruno, "Schrödinger Filtering: A Novel Technique for Removing Gradient Artifact from Electroencephalography Data Acquired during Functional Magnetic Resonance Imaging" (2019). Electronic Thesis and Dissertation Repository. 6606.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 4.0 License.