TubeTK/Events/2011.07.06: Difference between revisions

From KitwarePublic
< TubeTK‎ | Events
Jump to navigationJump to search
No edit summary
No edit summary
Line 1: Line 1:
UCLA & Kitware meeting notes:
UCLA (Andrei) & Kitware (Stephen, Gabe, Daniel, Ilknur) meeting notes:


== Segmentation in the chronic image ==
== Segmentation in the chronic image ==

Revision as of 13:48, 7 July 2011

UCLA (Andrei) & Kitware (Stephen, Gabe, Daniel, Ilknur) meeting notes:

Segmentation in the chronic image

Motivation:

Tracking tissue fate in TBI is currently a poorly understood problem with significant clinical implications. One such implication is the ability to obtain highly accurate (A) WM/GM tissue classification at acute baseline as well as (B) metrics of atrophy between the acute and chronic time points. Currently this can be very difficult. Another implication is the ability to measure atrophy with accuracy at the location of the primary injury, which can also be challenging.

Problem:

Tissue classification is available at the chronic time point. Because of diadema, tracking of tissues and classification of white and gray matters at the acquired time point is hard. In acute phase, lesion boundaries should be hand segmented. How to label a voxel based on 5 modalities is a problem.

Possible approach:

With the classification in the chronic image, acute image and Gabe&Marc`s algorithm, can this (segmentation) be done in the acute image?

Kitware can do multivariant classification. PDFSegmenter is available in TubeTK. This method is semiautomatic, multivariant, and multiclass (2 types of diadema & tumor). In this approach, you should indicate couple of points in diadema and some other points. Andrei told that Marcel has something similar to that one. Both can be tried and the results can be compared.

Questions:

Whether a diadema is hemorrhagic or non-hemorrhagic cannot be generalized. Micro bleeds happens and they are visible in SWI (not visible in the others).


Input:

  1. Data: At least 4-5 channels (T1, T2, SWI, DTI, GRE) are available. They are good for differenting diademas.
  2. Lesion classes: How to distinguish between lesion classes? Andrei has a submitted paper about that.

Brain inflammation at 3D volume (Deformation due to swelling)

Motivation:

Understanding how the brain swells during TBI is very important clinically because much of TBI critical care must address the phenomenon of brain inflammation, either pharmacologically or otherwise. Furthermore, TBI inflammation mechanisms are not understood as well as might be desired.

Problem:

How brain changes its shape due to swelling (Inflammation of brain)? Can the approach in Gabe and Marc`s paper be applied to solve that problem?

Possible solution:

There is a filter called, JacobianDeformationField, in ITK. The challenge here is how do we interpolate in brain region which is consistently swelling or shrinking. Localization of the swelling is important (Swelling will be gone in the chronic image). Example: When the primary injury is small and there is a lot of swelling.

Questions:

  1. How brain swells? How brain inflamms as a result of injury? How gray&white matter gets effected by that? These are important.
  2. How swelling benefits the injury parts? (Ventriculoscopy pressure, inflammation)
  3. What are the measures that tell us there is swelling? If gray&white matter segmentation and labelling diadema gives information about that, we can do that.

Vocabulary: Ventriculoscopy pressure: [[1]]


Next Step

  1. Andrei will give us the manual segmentation of the subject
  2. We will do segmentation for diadema (hemorrhage or not).
  3. Marcel and Guido`s segmentation method can also be applied.