Fast diffusion MRI based on sparse acquisition and reconstruction for long-term population imaging

Dissertation, University of Bonn, Nov. 2019
 

Abstract

Diffusion weighted magnetic resonance imaging (dMRI) is a unique MRI modality to probe the diffusive molecular transport in biological tissue. Due to its noninvasiveness and its ability to investigate the living human brain at submillimeter scale, dMRI is frequently performed in clinical and biomedical research to study the brain’s complex microstructural architecture. Over the last decades large prospective cohort studies have been set up with the aim to gain new insights into the development and progression of brain diseases across the life span and to discover biomarkers for disease prediction and potentially prevention. To allow for diverse brain imaging using different MRI modalities, stringent scan time limits are typically imposed in population imaging. Nevertheless, population studies aim to apply advanced and thereby time consuming dMRI protocols that deliver high quality data with great potential for future analysis. To allow for time-efficient but also versatile diffusion imaging, this thesis contributes to the investigation of accelerating diffusion spectrum imaging (DSI), an advanced dMRI technique that acquires imaging data with high intra-voxel resolution of tissue microstructure. Combining state-of-the-art parallel imaging and the theory of compressed sensing (CS) enables the acceleration of spatial encoding and diffusion encoding in dMRI. In this way, the otherwise long acquisition times in DSI can be reduced significantly. In this thesis, first, suitable q-space sampling strategies and basis functions are explored that fulfill the requirements of CS theory for accurate sparse DSI reconstruction. Novel 3D q-space sample distributions are investigated for CS-DSI. Moreover, conventional CS-DSI based on the discrete Fourier transform is compared for the first time to CS-DSI based on the continuous SHORE (simple harmonic oscillator based reconstruction and estimation) basis functions. Based on these findings, a CS-DSI protocol is proposed for application in a prospective cohort study, the Rhineland Study. A pilot study was designed and conducted to evaluate the CS-DSI protocol in comparison with state-of-the-art 3-shell dMRI and dedicated protocols for diffusion tensor imaging (DTI) and for the combined hindered and restricted model of diffusion (CHARMED). Population imaging requires processing techniques preferably with low computational cost to process and analyze the acquired big data within a reasonable time frame. Therefore, a pipeline for automated processing of CS-DSI acquisitions was implemented including both in-house developed and existing state-of-the-art processing tools. The last contribution of this thesis is a novel method for automatic detection and imputation of signal dropout due to fast bulk motion during the diffusion encoding in dMRI. Subject motion is a common source of artifacts, especially when conducting clinical or population studies with children, the elderly or patients. Related artifacts degrade image quality and adversely affect data analysis. It is, thus, highly desired to detect and then exclude or potentially impute defective measurements prior to dMRI analysis. Our proposed method applies dMRI signal modeling in the SHORE basis and determines outliers based on the weighted model residuals. Signal imputation reconstructs corrupted and therefore discarded measurements from the sparse set of inliers. This approach allows for fast and robust correction of imaging artifacts in dMRI which is essential to estimate accurate and precise model parameters that reflect the diffusive transport of water molecules and the underlying microstructural environment in brain tissue.

Download: https://nbn-resolving.org/urn:nbn:de:hbz:5n-56265

Bibtex

@PHDTHESIS{koch2019phd,
    author = {Koch, Alexandra},
     title = {Fast diffusion MRI based on sparse acquisition and reconstruction for long-term population imaging},
      type = {Dissertation},
      year = {2019},
     month = nov,
    school = {University of Bonn},
  abstract = {Diffusion weighted magnetic resonance imaging (dMRI) is a unique MRI modality to probe the diffusive
              molecular transport in biological tissue. Due to its noninvasiveness and its ability to investigate
              the living human brain at submillimeter scale, dMRI is frequently performed in clinical and
              biomedical research to study the brain’s complex microstructural architecture. Over the last
              decades large prospective cohort studies have been set up with the aim to gain new insights into the
              development and progression of brain diseases across the life span and to discover biomarkers for
              disease prediction and potentially prevention. To allow for diverse brain imaging using different
              MRI modalities, stringent scan time limits are typically imposed in population imaging.
              Nevertheless, population studies aim to apply advanced and thereby time consuming dMRI protocols
              that deliver high quality data with great potential for future analysis.
              To allow for time-efficient but also versatile diffusion imaging, this thesis contributes to the
              investigation of accelerating diffusion spectrum imaging (DSI), an advanced dMRI technique that
              acquires imaging data with high intra-voxel resolution of tissue microstructure. Combining
              state-of-the-art parallel imaging and the theory of compressed sensing (CS) enables the acceleration
              of spatial encoding and diffusion encoding in dMRI. In this way, the otherwise long acquisition
              times in DSI can be reduced significantly.
              In this thesis, first, suitable q-space sampling strategies and basis functions are explored that
              fulfill the requirements of CS theory for accurate sparse DSI reconstruction. Novel 3D q-space
              sample distributions are investigated for CS-DSI. Moreover, conventional CS-DSI based on the
              discrete Fourier transform is compared for the first time to CS-DSI based on the continuous SHORE
              (simple harmonic oscillator based reconstruction and estimation) basis functions.
              Based on these findings, a CS-DSI protocol is proposed for application in a prospective cohort
              study, the Rhineland Study. A pilot study was designed and conducted to evaluate the CS-DSI protocol
              in comparison with state-of-the-art 3-shell dMRI and dedicated protocols for diffusion tensor
              imaging (DTI) and for the combined hindered and restricted model of diffusion (CHARMED). Population
              imaging requires processing techniques preferably with low computational cost to process and analyze
              the acquired big data within a reasonable time frame. Therefore, a pipeline for automated processing
              of CS-DSI acquisitions was implemented including both in-house developed and existing
              state-of-the-art processing tools.
              The last contribution of this thesis is a novel method for automatic detection and imputation of
              signal dropout due to fast bulk motion during the diffusion encoding in dMRI. Subject motion is a
              common source of artifacts, especially when conducting clinical or population studies with children,
              the elderly or patients. Related artifacts degrade image quality and adversely affect data analysis.
              It is, thus, highly desired to detect and then exclude or potentially impute defective measurements
              prior to dMRI analysis. Our proposed method applies dMRI signal modeling in the SHORE basis and
              determines outliers based on the weighted model residuals. Signal imputation reconstructs corrupted
              and therefore discarded measurements from the sparse set of inliers. This approach allows for fast
              and robust correction of imaging artifacts in dMRI which is essential to estimate accurate and
              precise model parameters that reflect the diffusive transport of water molecules and the underlying
              microstructural environment in brain tissue.},
       url = {https://nbn-resolving.org/urn:nbn:de:hbz:5n-56265}
}