Upload
jaimie
View
44
Download
0
Tags:
Embed Size (px)
DESCRIPTION
P025 MPRAGE Pre-Contrast. P025 MPRAGE w/ Z-Score < -4. Notes. This was accomplished by doing an inverse nonlinear warp from MNI to the SPGR FA 18, then to the MPRAGE space with a linear transform (ideally these transforms should be combined) - PowerPoint PPT Presentation
Citation preview
P025 MPRAGE Pre-Contrast
P025 MPRAGE w/ Z-Score < -4
Notes
• This was accomplished by doing an inverse nonlinear warp from MNI to the SPGR FA 18, then to the MPRAGE space with a linear transform (ideally these transforms should be combined)
• Images are larger than shown, zoom in for more detail• Not all lesions are found by the z-score thresholding but
a significant amount are• Large volumes of white matter are found to be
significantly demyelinated: perhaps DAWM or actually revealing the “invisible disease”?
• This was one the patients with a larger amount of low MWF volume, EDSS 5.0 SPMS
Another Look at Multi-Channel Segmentation with FSL FAST
MPRAGE
• Same slices as shown in the z-score slides• Lesions of interest circled
SPGR 3 class-CSF
• Generally good CSF segmentation• Does not capture most lesions
SPGR 3 class-GM
• Includes some or all of lesions
SPGR 3 class-WM
• Matches nicely with the MPRAGE scan• Partially includes lesions
SPGR 4 class-CSF
• Generally worse than 3 class CSF
SPGR 4 class-GM
• More conservative estimate of GM, much fewer lesions included
SPGR 4 class-More GM
• Deep GM• Includes many of the lesions
SPGR 4 class-WM
• Does well at excluding most focal lesions but appears to be some partially including some
• More conservative
SPGR-FLAIR 3 class-CSF
• Grabs most lesions• Unfortunately mis-classifies GM too
SPGR-FLAIR 3 class-GM
• Includes many regions previously seen as WM
SPGR-FLAIR 3 class-WM
• Good exclusion of lesions identified by z-score• Does poorly at WM and GM segmentation,
misses some WM
SPGR-FLAIR 4 class-CSF
• Generally worse than 3 class CSF also• Chokes back mask too far
SPGR-FLAIR 4 class-GM
• Still misidentifies a lot of WM as GM• Catches edges of our lesions of interest
SPGR-FLAIR 4 class-More GM
• Again mis-includes a lot of WM
SPGR-FLAIR 4 class-WM
• Avoids lesions but also misses a lot of regular WM since those are misidentified as GM
SPGR-T2-PD 3 class-CSF
• Decent at outside brain CSF, though catches some WM
• Does not get any CSF inside brain
SPGR-T2-PD 3 class-GM
• Captures our lesions of interest• Gets ventricle CSF• Not as good as SPGR 3 class
SPGR-T2-PD 3 class-WM
• Overly greedy WM• Misses focal lesions but gets GM
SPGR-T2-PD 4 class-CSF
• Same problems as with SPGR-T2-PD CSF segmentation
SPGR-T2-PD 4 class-GM
• GM + inner CSF• Catches darker lesions• Pretty poor, also gets WM
SPGR-T2-PD 4 class-More GM
• Catches some lesions but overall pretty garbagey
• Doesn’t really correspond to a distinct tissue class
SPGR-T2-PD 4 class-WM
• A conservative estimate• Circled region shows possible inclusion of GM
and missing of brain stem
Best CSF
• SPGR 3 class (includes some lesions)
• SPGR-FLAIR 3 class (includes all lesions)
• SPGR-T2-PD 3 class (useful for out of brain CSF, does not include lesions)
Best WM
• SPGR 3 class (best anatomically)• SPGR 4 class (conservative)• SPGR-FLAIR 4 class (more lesion exclusion)
Best GM
• SPGR 3 class (GM+lesions)
• SPGR 4 class (deep GM+lesions)
Best Lesion
• SPGR 3 class (GM+lesions)
• SPGR-FLAIR 3 class (CSF missing some lesions)
Coming Soon
• Segmentation with a priori maps
• Ideas about how to combine maps to produce NAWM, NAGM, and lesion only masks