ISMRM 23rd Annual Meeting & Exhibition • 30 May - 05 June 2015 • Toronto, Ontario, Canada

Scientific Session • Diffusion Weighted Image Analyses

Friday 5 June 2015

Room 718 A

08:00 - 10:00


Andrew L. Alexander, Ph.D., Chantal M. W. Tax, M.Sc.

08:00 1023.   
Noise map estimation in diffusion MRI using Random Matrix Theory
Jelle Veraart1, Els Fieremans1, and Dmitry S. Novikov1
1Center for Biomedical Imaging, NYU Langone Medical Center, New York, NY, United States

We propose a new technique to estimate the spatially varying noise map based on diffusion MRI data to enable Rician bias correction. The technique makes use of a random matrix theorem, i.e. Marchenko-Pastor’s law, to estimate the noise level by exploiting the redundancy in multi-directional diffusion MR data.

08:12 1024.   Caveats of non-linear fitting to brain tissue models of diffusion
Ileana O. Jelescu1, Jelle Veraart1, Els Fieremans1, and Dmitry S. Novikov1
1Center for Biomedical Imaging, Dept. of Radiology, NYU Langone Medical Center, New York, New York, United States

Compared to DTI, white/gray matter models of diffusion should have improved specificity. However, fit outputs notoriously suffer from bias and poor precision, with most models employing simplifying assumptions to stabilize the fit. Here, we use the example of NODDI to assess the behavior of nonlinear fitting when all model parameters are free. We reveal that the typical full model of brain tissue cannot be reliably determined, due to a duality of solutions, and to the narrow and shallow (boomerang-shaped) minimization landscape. Constraining the fit with fixed parameter values that lack biological validation is not a trustworthy solution to the problem.

08:24 1025.   
Joint estimation of microstructural and biomechanical features of the brain using a phase sensitive reconstruction of DWIs
Tim Sprenger1,2, Jonathan I. Sperl2, Axel Haase1, Brice Fernandez3, Christopher Hardy4, Luca Marinelli4, Michael Czisch5, Philipp Saemann5, and Marion I. Menzel2
1IMETUM, Technical University, Munich, Germany, 2GE Global Research, Munich, Germany, 3GE Healthcare, Munich, Germany, 4GE Global Research, Niskayuna, NY, United States, 5Max Planck Institute of Psychiatry, Munich, Select, Germany

Diffusion weighted magnetic resonance imaging (DWI) allows for non-invasive measurement of microstructural features of the human brain. Usually all data processing in DWI is based on the magnitude of the complex MR signal, and the inherent phase of the signal is discarded as it is considered to be spoiled by different sources. One source for a non-zero phase in DWI signal is the pulsation of the brain itself, which is potentially impacted by disorders such as hydrocephalus or traumatic brain injury. In this work, a phase correction approach is introduced to sequentially remove all phase contributions except for brain pulsation.

08:36 1026.   A compressed sensing approach to super-resolution diffusion MRI from multiple low-resolution images
Lipeng Ning1,2, Kawin Setsompop2,3, Cornelius Eichner3, Oleg Michailovich4, Carl-Fredrik Westin1,2, and Yogesh Rathi1,2
1Brigham and Women's Hospital, Boston, MA, United States, 2Harvard Medical School, Boston, MA, United States, 3Massachusetts General Hospital, MA, United States, 4University of Waterloo, Ontario, Canada

We present a novel compressed sensing approach for super resolution reconstruction (SRR) of diffusion MRI using multiple anisotropic low-resolution images. The diffusion signal in each voxel is estimated using spherical ridgelets while the spatial correlation between neighboring voxels is accounted for using total-variation (TV) regularization. The experimental result using in-vivo human brain data shows that the proposed SRR method is capable of recovering complex fiber orientations at a very high spatial resolution, similar to a physically acquired “gold-standard” data. Hence it has potential to be applied in clinical settings to study mental diseases and to reduce partial-volume effect.

08:48 1027.   
Time to move on: an FOD-based DEC map to replace DTI's trademark DEC FA
Thijs Dhollander1, Robert Elton Smith1, Jacques-Donald Tournier2,3, Ben Jeurissen4, and Alan Connelly1,5
1The Florey Institute of Neuroscience and Mental Health, Melbourne, Victoria, Australia, 2Centre for the Developing Brain, King's College London, London, United Kingdom, 3Department of Biomedical Engineering, King's College London, London, United Kingdom, 4iMinds-Vision Lab, University of Antwerp, Antwerp, Belgium, 5The Florey Department of Neuroscience, University of Melbourne, Melbourne, Victoria, Australia

The "traditional" directionally-encoded colour (DEC) FA map is an icon of DTI, but is also affected by its inherent flaws. The first eigenvector is known to be ill-defined in regions of crossing fibres, resulting in misleading specific DEC values as well as "false edges" in the overall map. Additionally, the FA shows naturally low values in these regions. In a clinical setting, this might potentially lead to false positive findings; but also to false negative ones in case these false features mask out or otherwise distract from real pathological features. We propose an FOD-based DEC map that solves these issues.

09:00 1028.   Resolving crossing fibers and generalizing biomarkers using the diffusion kurtosis tensor
Rafael Neto Henriques1, Marta Morgado Correia1, Rita Gouveia Nunes2, and Hugo Alexandre Ferreira2
1Cognition and Brain Science Unit, MRC, Cambridge, England, United Kingdom, 2Instituto de Biofisica e Engenharia Biomedica, Faculdade de Ciencias da Universidade de Lisboa, Lisbon, Lisbon, Portugal

Diffusion Kurtosis Imaging (DKI) models the non-Gaussian behaviour of water diffusion by the diffusion kurtosis tensor (KT), which can be used to provide indices of tissue heterogeneity and a better characterisation of the spatial architecture of tissue microstructure. In this study, the advantages and disadvantages of using KT based fiber direction estimates to resolve crossing fibers and compute a generalized version of radial kurtosis (RK) are investigated. Our results show that KT fiber direction estimates provides smaller angular errors when compared to previous DKI fiber estimation procedures and RK measures less sensitive to noise bias.

09:12 1029.   
Comparing Fourier to SHORE Basis Functions for Sparse DSI Reconstruction
Alexandra Tobisch1,2, Thomas Schultz2, Rüdiger Stirnberg1, Gabriel Varela3, Hans Knutsson4, Pablo Irarrázaval3,5, and Tony Stöcker1,6
1German Center for Neurodegenerative Diseases, Bonn, Germany, 2Department of Computer Science, University of Bonn, Bonn, Germany, 3Biomedical Imaging Center, Pontificia Universidad Católica de Chile, Santiago, Chile, 4Linköping University, Linköping, Sweden, 5Department of Electrical Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile, 6Department of Physics and Astronomy, University of Bonn, Bonn, Germany

Compressed Sensing (CS) theory accelerates Diffusion Spectrum Imaging (DSI) acquisition, while still providing high angular and radial resolution of intra-voxel microstructure. Several groups have proposed to reconstruct the diffusion propagator from sparse q-space samples by fitting continuous basis functions. Among these, the SHORE basis has recently been found to perform best. This work compares the SHORE-based approach to traditional CS recovery that combines the discrete Fourier transform with a sparsity term. For simulated diffusion signals, the CS reconstruction is found to deviate less from the ground truth when using Fourier basis functions for sparse DSI reconstruction.

09:24 1030.   How to avoid biased streamlines-based metrics for streamlines with variable step sizes
Jean-Christophe Houde1, Marc-Alexandre Côté-Harnois1, and Maxime Descoteaux1
1Computer Science department, Université de Sherbrooke, Sherbrooke, Quebec, Canada

We show that metrics computed over streamlines can easily be biased or incorrect for streamlines with a step size that is too large or variable. The basic methods to compute those statistics, sometimes called Tractometry methods, generally only use the points of the streamlines to sample the corresponding image volumes. However, for streamlines where the step size is too large or variable, this sampling is skewed, and derived metrics are biased. We present a simple updated method that correctly handles those streamlines, and we show that metrics computed using this method are robust to the streamline sampling.

09:36 1031.   
Imposing label priors in global tractography can resolve crossing fibre ambiguities
Daan Christiaens1,2, Frederik Maes1,2, Stefan Sunaert2,3, and Paul Suetens1,2
1Electrical Engineering, KU Leuven, Leuven, Vlaams-Brabant, Belgium, 2Medical Imaging Research Center, UZ Leuven, Leuven, Vlaams-Brabant, Belgium,3Translational MRI, KU Leuven, Leuven, Vlaams-Brabant, Belgium

Ambiguity in the local diffusion profile is an important cause of spurious fibre tracks in DWI. We propose to use bundle labels as an additional prior in global tractography to overcome this issue. We have evaluated such label prior in a Tractometer study, and demonstrated its effect on real data using the Catani atlases of 30 white matter tracts. Results show an important decrease of invalid connections and spurious tracks. Additionally, a probabilistic segmentation of the fibre bundles is obtained.

09:48 1032.   Connectivity based segmentation of the Corpus Callosum using a novel data mining approach
Gowtham Atluri1, An Wu2, Essa Yacoub2, Kamil Ugurbil2, Vipin Kumar1, and Christophe Lenglet2
1Computer Science and Engineering, University of Minnesota, Minneapolis, Minnesota, United States, 2Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, Minnesota, United States

Existing approaches that are used to do finer segmentation of cortical regions using DTI based tractography data do not make use of the underlying spatial structure. In this work, we extend a popular Shared Nearest Neighbor (SNN) based clustering approach in order to account for spatial structure in the data. We use this approach to discover finer segmentation of the Corpus Callosum in 2 normal subjects using tractography data computed from a 3T and a 7T DTI scan. Our results suggest that our new approach results in a segmentation that is consistent between 3T and 7T.