Computational Anatomy

We develop algorithms for automated structural alignment, necessary for large-scale studies of diffusion MRI data. These algorithms cater to various diffusion models, such as the diffusion tensor, spherical harmonics-based models, and multi-tensor models. We use deformation models ranging from small elastic deformation to large flow-based deformation. Various analyses can be performed after registration: 1) Voxel-based morphometry (VBM), where the spatially aligned images are compared statistically voxel-by-voxel; 2) Deformation-based morphometry (DBM), where shape variations are evaluated via analysis of the deformation fields.

Image Enhancement

Diffusion MRI data suffer from relatively low spatial resolution and signal-to-noise ratio due to the long echo time (TE) requirement typical of an echo-planar imaging (EPI) sequence. We develop algorithms for improving the quality diffusion MRI data via biologically motivated interpolation techniques, maximizing the amount of information that can be extracted from the acquired data. We can typically up-sample the data via a 64 times reduction in voxel volume, significantly increasing structural clarify for a closer inspection of white matter abnormalities.

  • P.-T. Yap, D. Shen, "Resolution Enhancement of Diffusion-Weighted Images by Local Fiber Profiling," MICCAI 2012, Nice, France, Oct. 1-5, 2012.
  • P.-T. Yap, H. An, Y. Chen, D. Shen, “Fiber-Driven Resolution Enhancement of Diffusion-Weighted Images - An Evaluation Using High Resolution Data,” ISMRM’13, Salt Lake City, Utah, USA, April 20-26, 2013.
  • P.-T. Yap, H. An, Y. Chen, D. Shen, "A Generative Model for Resolution Enhancement of Diffusion MRI Data," MICCAI 2013, Nagoya, Japan, Sep. 22-26, 2013.
  • F. Shi, J. Cheng, L. Wang, P.-T. Yap, D. Shen, "Low-Rank Total Variation for Image Super Resolution," MICCAI 2013, Nagoya, Japan, Sep. 22-26, 2013.

Diffusion Models

We have extensive experience in employing constrained optimization techniques (involving sparsity, non-negativity, unit integral PDF constraint, total variation, Riemannian manifold, neighborhood consistency, anatomical constraint, non-locality, and representation learning) for robust parameter estimation of diffusion models that will allow us to compute the fiber/diffusion orientation distribution function and the ensemble average propagator.

  • J. Cheng, D. Shen, P.-T. Yap, “Non-Negative Spherical Deconvolution for Fiber Orientation Distribution Estimation,” ISMRM’13, Salt Lake City, Utah, USA, April 20-26, 2013.
  • J. Cheng, R. Deriche, T. Jiang, D. Shen, P.-T. Yap, “Non-Local Non-Negative Spherical Deconvolu- tion for Single and Multiple Shell Diffusion MRI,” HARDI Reconstruction Challenge, International Symposium on Biomedical Imaging, 2013. (Honorable Mention; best technique in terms of local comparison)
  • P.-T. Yap, D. Shen, "Spatial Transformation of DWI Data Using Non-Negative Sparse Representation ," IEEE Transactions on Medical Imaging, vol. 31, no. 11, pp. 2035-2049, 2012.

Compressed Sensing

We develop algorithms for accelerating the acquisition of diffusion data by employing sub-Nyquist reconstruction techniques, reducing significantly the required number of signal measurements. Our techniques allow acquisition of data with higher spatial resolution and greater angular contrast in a lesser amount of time.

  • J. Cheng, T. Jiang, R. Deriche, D. Shen, P.-T. Yap, "Regularized Spherical Polar Fourier Diffusion MRI with Optimal Dictionary Learning," MICCAI 2013, Nagoya, Japan, Sep. 22-26, 2013.


We develop algorithms for estimating trajectories of axonal fascicles involving a single subject, a population of subjects, and a single subject scanned longitudinally.


Tractography data can be utilized to quantify region-to-region connectivity, which can be in turn used for a graph-theoretic analysis of brain circuitry in relation to brain growth, aging, and diseases. This approach has been applied for investigation of Alzheimer's disease, schizophrenia, multiple sclerosis, and brain development in the first years of life.