Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article

Degree

Doctor of Philosophy

Program

Medical Biophysics

Supervisor

Drangova, Maria

Abstract

Magnetic resonance imaging (MRI) is an invaluable medical imaging modality providing excellent soft tissue contrast for diagnostic and research purposes. However, the long scan times required for image acquisition increase the possibility for subject motion, with the resulting artifacts potentially impeding diagnosis or other analysis. Even for brain imaging, where involuntary motion (e.g. breathing) is not expected to greatly influence acquisition, it is estimated that one in five MRI exams require at least one scan be repeated due to motion artifacts.

This thesis is centered on the development of motion correction methods for brain MRI. The first objective was to extend spherical navigators (SNAVs), a form of MR-based motion tracking previously validated for retrospective motion correction, to more advantageous prospective applications. The next objective was to demonstrate the first purely navigator-based approach to prospective motion correction (PMC) for R2* and susceptibility mapping. The last objective, for cases where retrospective correction is preferable, was to improve deep-learning-based motion correction by leveraging the additional degree of spatial encoding embedded in multichannel data.

Results demonstrated that optimized processing of SNAVs for real-time application enables PMC of brain MRI with sub millimeter accuracy and low latency (within 60 MS). Furthermore, implementing SNAV-based PMC in a multi-echo gradient echo protocol for R2* and susceptibility mapping in the presence of motion was shown to consistently reduce quantitative mapping error, further improved with retrospective correction of motion-induced field offsets. Finally, in the context of deep-learning-based correction, results showed a 40% improvement in error reduction when applying the networks prior to coil combination.

In summary, this thesis presents three notable developments in motion correction for brain MRI. Improved motion correction methods could reduce the need for repeat scans, which increase costs and can delay diagnosis in a clinical setting.

Summary for Lay Audience

Magnetic resonance imaging (MRI) is a very useful tool for medical diagnosis and research improving our understanding of brain development, aging, and disease. However, a single MRI can take several minutes, making it hard for the person being scanned to hold still for the process. Even small motions of a few millimeters can result in image blurring and other artifacts which make it difficult for doctors to make a diagnosis or perform other analyses.

This thesis focuses on further developing ways to correct for motion during brain MRIs. One approach, denoted prospective motion correction (PMC), updates the MRI scan in real-time to account for motion, preventing artifacts. To inform these updates, the MRI scanner itself can track head motion using navigators acquired throughout the scan. Alternatively, motion may be corrected retrospectively, in which case corrupted images are processed to remove artifacts after the scan is complete. For example, artificial intelligence (AI) may be used to train very complex models to recognize and remove motion artifacts from corrupted images.

The first goal of this work was to demonstrate PMC using a specific MRI tracking method denoted spherical navigators (SNAVs). Results showed, for the first time, that SNAVs can be used for real-time tracking of head motion. Next, it aimed to apply SNAV-based PMC to quantitative mapping, an MRI technique measuring specific tissue properties to help identify disease. The proposed correction technique was shown to improve the accuracy of these measurements in the presence of head motion. The final goal was to improve AI-based motion correction by making better use of multichannel MRI data. This multichannel data consists of many images obtained during an MRI scan, each sensitive to different brain regions, later combined to form the final image. It was determined that AI models correcting these individual channel images more effectively remove artifacts compared to similar AI models applied after channel combination.

If improved motion correction methods were applied in a hospital setting, they could reduce the need to repeat patient scans, lowering healthcare costs and preventing delays in diagnosis. MRI wait times could also be improved through increased scanning efficiency.

Creative Commons License

Creative Commons Attribution-Noncommercial 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License

Download

Share

COinS