1. Research Update
Nov 18, 2016
Jason Su

2. 3T Pre-MS In collaboration with Bordeaux (recruiting, scanning)
23 CIS patients, 16 controls
Thomas Tourdias, Fanny Munsch, Sandy Mournet
Study thalamic atrophy in CIS/early MS and relation to cognitive performance
Detailed neuropsych testing
Verbal memory learning and recall
Visual memory learning and recall
Working memory
Executive function
Information processing speed

3. Imaging and Post-Processing
Acquiring CSFnMPRAGE (WM/GM structural), FLAIR (lesions), DTI, WMnMPRAGE (thalamic anatomy)
Primarily been working with the WMnMPRAGE images
Brain parenchyma and intracranial cavity (ICC) segmentation
Thalamic anatomy segmentation (12 nuclei and whole thalamus)

4. Segmentation Thalamus Brain and Intra-Cranial Cavity
Fanny sends images
Jason applies THOMAS (v0, whole brain template variant), ~5-6h per brain per side
Sandy edits segmentations with training from Thomas
A label-fusion based algorithm to segment these quantities
Priors derived from multi-contrast co-registered volumes from ET-FUS study
Brain parenchyma priors from FSL’s BET of CSFnMPRAGE
ICC priors from BET of CUBE
Co-registered onto 3T WMnMPRAGE
Majority voting with some smoothing
Another WMn template specific to the FUS study, ~5-6h per brain
Sandy edits ICC as well
Should we define ICV (intra cranial volume) and TPV (total parenchymal volume) based on editing down to supra-tentorial only, as we / Hagen did in MSmcDESPOT?

5. Statistical Analysis
Must be very careful about multiple comparisons problem
2x13 thalamic regions
10 cognitive scores
Only 21 patients
Want to narrow down the number of correlations we test to a few select high-value targets
Anterior thalamus and memory-guided attention
MD and learning and decision-making
Narrow down which thalamic regions to study in multivariate analysis, by comparing patient group to control group
We expect only those regions that are significantly altered from normal to be correlated to the disease state
I believe we only have 20 complete patient datasets right now, isn’t that right? Because P057 had thalamic lesions so dropped out of the patient group?
My master spreadsheet shows only 13 controls: T 001 to T 013.
Question: based on the way the THOMAS template was created from single-sided manual segmentations of MS+HC brains, could this introduce any kind of net “bias” or offset in right vs left thalamic volumes for incoming brains? This could be something to be very careful with, and how we describe / justify this in the paper
Mitchell references:
Mitchell et al FrontSysNeurosci What does the mediodorsal thalamus do?
Mitchell et al NeurosciBiobehavioralRev The mediodorsal thalamus as a higher order thalamic relay nucleus important for learning and decision making

6. Group Comparisons: Raw THOMAS ICV “regressed out”

7. Group Comparisons: Raw THOMAS ICV “regressed out”

8. Group Comparisons: Raw THOMAS ICV “regressed out”

9. Group Comparisons: Raw THOMAS ICV “regressed out”

10. Memory ~ Age + Sex + SNR + ICV + Mean_Thalamus + Mean_AV
Adjusts for gender, SNR, age, ICV, whole thal

11. Memory ~ Age + Sex + SNR + ICV + Mean_Thalamus + Mean_AV
Adjusts for gender, SNR, age, ICV, whole thal

12. Decision ~ Age + Sex + SNR + ICV + Mean_Thalamus + Mean_MDPf
Adjusts for gender, SNR, age, ICV, whole thal

13. Decision ~ Age + Sex + SNR + ICV + Mean_Thalamus + Mean_MDPf
Adjusts for gender, SNR, age, ICV, whole thal

14. Group Comparisons: Raw vs Edited ICV “regressed out”

15. Group Comparisons: Raw vs Edited ICV “regressed out”

16. Group Comparisons: Raw vs Edited ICV “regressed out”

17. Simple Correlation
Other papers use simple correlations as a way to build multiple regression model in steps (only use covariates that are significant)
However, there are critics of this approach
Since our patient population is pretty uniform, we remove the 3 males and only regress out ICV

18. Simple Correlations (Spearman)

19. Simple Correlations (Spearman)

20. Decision ~ Age + Sex + SNR + ICV + Mean_Thalamus + Mean_VPL

21. Decision ~ Age + Sex + SNR + ICV + Mean_Thalamus + Mean_VPL

22. Future work
Lesion segmentation (if any?)
WM/GM atrophy?
VBM/TBM/DBM