Difference between revisions of "2012 Winter Project Week:GBMseg"
| (2 intermediate revisions by the same user not shown) | |||
| Line 15: | Line 15: | ||
<h3>Approach, Plan</h3> | <h3>Approach, Plan</h3> | ||
| − | The approach is to obtain accurate registration of a number of MR sequences of each tumor case, and analyze correlation with pathology data. | + | The approach is to obtain accurate registration of a number of MR sequences of each tumor case, segment what is thought to be the tumor on T1 post contrast and analyze correlation with pathology data. |
</div> | </div> | ||
| Line 22: | Line 22: | ||
<h3>Progress</h3> | <h3>Progress</h3> | ||
| + | |||
| + | Segmented one T2 FLAIR image (W009), voxel size 1x1x5mm using growcut region algorithm. Manual pre-labeling was done on approximately central slice of the tumor in three section planes separately. Two-class tissue model (enhancing and non-enhancing region) was used. | ||
| + | Results: visually, the algorithm picked out most (about 80%) of the enhancing region. Some sharp mass processes were missed, as well as some areas that subjectively might not belong to tumor were also attributed to tumor. Comparing with hi-res T1 post-contrast (1x1x2mm voxel), revealed significant areas that highlight differently on T1 and T2, esp. around tumor boundary. Another accuracy limiting factor was high anisotropy of T2 acquisition, which resulted in ghosted data, particularly observabe in the areas of low signal, such as adjacent to bone. | ||
| + | |||
</div> | </div> | ||
</div> | </div> | ||
| Line 28: | Line 32: | ||
==Delivery Mechanism== | ==Delivery Mechanism== | ||
| + | GBM MR preprocessing pipeline | ||
==References== | ==References== | ||
Latest revision as of 17:21, 10 January 2012
Home < 2012 Winter Project Week:GBMsegContents
Key Investigators
- WUSM: Dan Marcus, Misha Milchenko
- Brigham and Women's: Andrey Fedorov, Jan Egger
Objective
We are looking to segment and visualize GBM's from various MR sequences in Slicer, and create an integrated display for clinician review, both radiologist and non-radiologist.
Approach, Plan
The approach is to obtain accurate registration of a number of MR sequences of each tumor case, segment what is thought to be the tumor on T1 post contrast and analyze correlation with pathology data.
Progress
Segmented one T2 FLAIR image (W009), voxel size 1x1x5mm using growcut region algorithm. Manual pre-labeling was done on approximately central slice of the tumor in three section planes separately. Two-class tissue model (enhancing and non-enhancing region) was used. Results: visually, the algorithm picked out most (about 80%) of the enhancing region. Some sharp mass processes were missed, as well as some areas that subjectively might not belong to tumor were also attributed to tumor. Comparing with hi-res T1 post-contrast (1x1x2mm voxel), revealed significant areas that highlight differently on T1 and T2, esp. around tumor boundary. Another accuracy limiting factor was high anisotropy of T2 acquisition, which resulted in ghosted data, particularly observabe in the areas of low signal, such as adjacent to bone.
Delivery Mechanism
GBM MR preprocessing pipeline
References
GrowCut segmentation Slicer documentation: http://wiki.slicer.org/slicerWiki/index.php/Modules:GrowCutSegmentation-Documentation-3.6
V. Vezhnevets and V. Konouchine, "GrowCut - Interactive multi-label N-D image segmentation," in Proc. Graphicon, 2005. pp. 150--156.
Mark Jenkinson, Stephen Smith. "Optimisation in Robust Linear Registration of Brain Images." FMRIB Technical Report TR00MJ2, http://www.fmrib.ox.ac.uk/analysis/techrep/tr00mj2/tr00mj2/
