Time Domain Principal Component Analysis for Rapid, Real-Time 2D MRI Reconstruction from Undersampled Data

• Author / Creator

  Wright, Mark Geoffrey

• Hybrid Linac-MR systems are becoming more mainstream for use in external beam radiotherapy treatments for cancer patients. The addition of an MR scanner to a linear accelerator (Linac) can allow for real-time structure tracking (such as a tumour) and on-the-fly adjustments of the radiation beam. This will allow for smaller treatment margins meaning more healthy tissue will be spared from irradiation. While MR provides superior soft-tissue contrast compared to other modalities, typical methods used are computationally expensive and require complicated setups or are otherwise too slow for real-time applications.
  The work proposed here utilizes a sliding window, Principal Component Analysis (PCA) in the temporal domain along with undersampled data from previously acquired frames in the sliding window to fully reconstruct the final frame within this window. Data is acquired such that a core set of phase encodes, located in central k-space, is acquired in every frame. PCA is performed on this data in order to characterize the temporal change in k-space. Outside of this core, the outer k-space is acquired in such a way that all of k-space is covered within a pre-determined number of frames. The principal components, which are continuously updated over an imaging session, are combined with the undersampled data to calculate weights which are used with the acquired data of the frame of interest to fill in the missing data.
  The method was tested retrospectively on 15 fully-sampled data sets of lung cancer patients acquired on a 3T MR scanner. A subset of six data sets was tested to determine the ability to contour a structure on the reconstructed images. The contours were developed using auto-contouring software and compared to contours developed on the original fully-sampled data sets.
  Due to changes in signal-to-noise ratio found at different MR field strengths, the algorithm was tested to determine the effects of added noise. Six data sets were used to retrospectively test the algorithm at noise levels of 2, 4 and 6 times greater than those calculated on the original 3T data. The reconstructed images were again tested for overall image quality as well as the ability to contour a structure on the reconstructed images.
  Previous work has utilized PCA corresponding to the spatial domain (intra-frame PCA method) to reconstruct images. The principal components were calculated from a database of fully-sampled images acquired just before the treatment. Initially the method worked very well, however as the imaging session progressed, image quality and contour metrics showed deterioration. Comparisons were made between this method and the one proposed in this work in order to test the robustness of the method. While the intra-frame method appeared to perform better initially, the through frame method maintained consistent metrics throughout an imaging session and performed better as time went on. This through frame PCA method appears to remain robust even at high acceleration rate and low SNR values.

• Subjects / Keywords

  • MRI
  • Real-time imaging
  • k-t data
  • PCA
  • Reconstruction

• Graduation date

  Spring 2022

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-8aay-bw35

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.