User:Jadrian Miles/Research Projects

This is a list of research projects, specific itemized aims for them, and potential groupings of aims into publications/presentations.

Scalar / Low-Dimensional Metrics

No concrete ideas on this so far, but I like the center-surround tract-transverse gradient concept. Gordon's presentation at ISMRM last year was neat; I wonder if he's cooked up anything new...

The connective coherence idea I had really seems to be the same thing as PICo, though the concept of normalizing relative to isotropic diffusion is still novel and could really be good.

FAR (Fibers At Risk) + PICo (Prob. Index of Connectivity)

Beefed-Up PICo

Aims:

1. Upgrade PICo (with directionality / merging fibers) to use a non-DT voxel model
   • Problem: PICo kinda smudges out the effect of the ODF anyway; would this gain us anything?
     • I guess that's research; you don't know until you try... it might follow branches (or not get confused at crossings) better
2. Run several PICo threads in true parallel on the GPU
   • Justification: faster PICo-based study turnarounds
3. Split PICo across a render farm
   • Justification: faster PICo-based study turnarounds

Pubs:

1. Robust PICo with RBFs
2. Fast PICo with a Graphics Render Farm

FAR

Aims:

1. Use longitudinal data and other contrasts to validate FAR
   • Justification: FAR was neat, but it needs validation
   • Justification: FAR may be a good tool for insight into MS mechanics and disease progression
2. Improve automatic lesion segmentation (maybe just by borrowing someone else's)
   • Justification: make FAR turnkey---more for validation studies than anything else, but would be great for anyone
3. Upgrade FAR to use PICo
   • Justification: improved longitudinal correlation
   • Justification: FAR + PICo may be an even better tool for studying MS

Pubs:

1. A Longitudinal Validation of Fibers-at-Risk Maps in Multiple Sclerosis
2. Computing Callosal Fibers-at-Risk Maps in Multiple Sclerosis with PICo

Whole-Brain Solving

As a spiral-developed concept, each of the following three sections should keep up with each other in parallel development.

WM Vector Graphics Model

Aims:

1. Define a simple, analytic description of volumetric WM tracts (additional parameters / refinements may be developed later)
2. "Thicken" a standard tractogram (or spin-glass tractogram?) to a vector graphics model; this is basically lossy compression and smoothing, so validate by re-generating the input tractogram
3. Use the higher-level information in the thickened model to bridge tractography gaps

Pubs:

1. An Analytic Model for White Matter Tract Structure

Compartment Model / Signal Simulation

Aims:

1. Simulate the diffusion MRI response, noise-free, for any combination of imaging parameters (including gradient direction) from a compartment model that includes some of the following:
   1. Multiple fiber directions in intersecting tracts
   2. Different axon calibers (or distributions of calibers) in different tracts
   3. Different isotropic D values for extracellular fluid and each tract's intracellular fluid
   4. Unequal volume fractions --- be careful that all the fibers and fluid can fit together!
   • Problem: but why does anybody care?
     • Demonstrate need by comparison to histology and just show different simulation results when using single calibers for all tracts vs. varied uniform calibers or distributions. When you consider multiple directions, even for the same b value, it may not be the case that the full signal profile for a distribution can also result from a simpler model.
   • Problem: how do you validate something with so many variables? It can't be solved for except with regularization or constraints.

Whole-Brain Solving

Aims:

1. Generate simulated diffusion images from a WM structural model (plain tractogram at first)
2. Adjust a WM structural model to better match its input images (DWIs and other contrasts)
   • Justification: must prove that standard tractography algorithms don't match their inputs well, but can be "nudged" to do better
3. Use global and tract-wise constraints (that obey biologically reasonable properties) to regularize this adjustment
4. Repeat everything with the analytic model

Pubs:

1. Post-Hoc Tractogram Adjustment Increases Conformance to Primary Images
2. Solving a Multi-Valued Analytic WM Structural Model with Post-Hoc Adjustment