A restriction-spectrum sparse-fascicle model for diffusion MRI

Ariel Rokem*, Christian Pötter, Robert F. Dougherty

Center for Cognitive and Neurobiological Imaging, Stanford University
*Now at the University of Washington eScience Institute

RS-SFM Combines two different ideas:

  • Sparse Fascicle Model (Rokem et al. 2015)

  • Restriction Spectrum Imaging (White et al. 2013)


    • Sparse Fascicle Model (Rokem et al., 2015)

  • The signal is modeled as a combination of tensor response functions:
    Response function Response function Response function Response function Response function ...

  • The active set of 'fascicles' is selected using non-negative least-squares regression regularized with Elastic Net
  • The direction-independent ("isotropic") signal is modeled separately as a multi-exponential: $$f(b) = \beta_1 e^{-bD_1} + \beta_2 e^{-bD_2} + \epsilon$$

    • Restriction Spectrum Imaging (White et al. 2013)

  • Each fascicle is repeated with different combinations of λ1,λ2,3
    • Response function Response function Response function Response function Response function Response function
  • Accounts for different restriction conditions within each voxel
  • A multi-exponential decay profile for every fascicle

    • Results (low gradient strength):

    But also (high gradient strength):

    $$Still, \: overall \: R^2 = 98.8$$

    Reproducible research

    "...an article about computational result is advertising, not scholarship. The actual scholarship is the full software environment, code and data, that produced the result..." -- Buckheit and Donoho (1995)

    All the code used to fit the model is available at

    http://arokem.github.io/ISBI2015

    Example analysis at: IPython nbviewer