External Email - Use Caution
Hi all,
I'm hoping to get some input on clustering, multiple correction, smoothing preferences, and (if available) best practices. Please correct me if I'm wrong, but it seems that the amount of attention given to multiple correction and spatial dependencies in fMRI hasn't yet translated to MRI/SBM. Doug & Bruce's recent paper (#1) beautifully highlights the value of permutation-based analyses for control of FPR (despite some of its complexities), but I'm wondering what (if any) consensus or reasoning exists as to the optimal spatial smoothing and correction methods that can be applied during permutations-based analysis of surface inputs (i.e., surface area, thickness, NPC of the two for volume, LGI, etc.).
1) The FSL page on randomise ( https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise/UserGuide) suggests that cluster-based thresholding based upon the null distribution of the maximum cluster size/extent is passé in comparison to using the maximum cluster mass. PALM offers both extent and mass options, but does this same sentiment for functional data apply to SBM?
2) My understanding is that LGI is already smoothed at ~ FWHM=15 during default processing of pial_lgi. If one were to look at area, thickness, and LGI, is there value to all variables being at the same level of smoothing (e.g., area @ 20mm, thickness @ 20mm, LGI @ 5mm)?
3) Whereas earlier papers (e.g., #2) advocated for p < .005 and cluster size > 10 voxels for functional analyses, for example, a number of folks have more recently suggested a high initial CFT (e.g., CFT=.001) and FWE = .05 (#s 3 & 4). Do any similar simulation results or suggestions exist for SBM rather than fMRI?
4) Does anyone know of guidance on FWE (or FDR or voxel/vertex-wise for that matter) thresholds for TFCE-corrected results?
5) Does anyone know of similar papers about FPRs in surface-based analyses that *also* consider false negative rates as well?
6) Have any journals made similar declarations about SBM/morphometry results like NeuroImage: Clinical did for functional data (#5)?
Although I'm neither the first nor the last person to ask these questions, I wanted to gather my thoughts and hopefully tap into the collective wisdom of this listserv community. My apologies if this email is seen as incendiary or faux pas, but I'm new to this world and would love any and all suggestions.
Thanks in advance for sharing any thoughts and helping rectify my naiveté! Dan
1: Greve, D. N., & Fischl, B. (2018). False positive rates in surface-based anatomical analysis. *NeuroImage*, *171*, 6-14. 2: Lieberman, M. D., & Cunningham, W. A. (2009). Type I and Type II error concerns in fMRI research: re-balancing the scale. *Social Cognitive and Affective Neuroscience, 4*(4), 423-428. 3: Eklund, A., Nichols, T. E., & Knutsson, H. (2016). Cluster failure: why fMRI inferences for spatial extent have inflated false-positive rates. PNAS, 113(28), 7900-7905. https://doi.org/10.1073/pnas.1602413113 4: Woo CW, Krishnan A, Wager TD (2014) Cluster-extent based thresholding in fMRI analyses: Pitfalls and recommendations. *NeuroImage, 91*, 412–419. 5: Roiser, J. P., Linden, D. E., Gorno-Tempinin, M. L., Moran, R. J., Dickerson, B. C., & Grafton, S. T. (2016). Minimum statistical standards for submissions to Neuroimage: Clinical. *NeuroImage: Clinical*, *12*, 1045.
On 2/4/19 7:26 PM, Daniel Leopold wrote:
External Email - Use Caution
Hi all,
I'm hoping to get some input on clustering, multiple correction, smoothing preferences, and (if available) best practices. Please correct me if I'm wrong, but it seems that the amount of attention given to multiple correction and spatial dependencies in fMRI hasn't yet translated to MRI/SBM. Doug & Bruce's recent paper (#1) beautifully highlights the value of permutation-based analyses for control of FPR (despite some of its complexities), but I'm wondering what (if any) consensus or reasoning exists as to the optimal spatial smoothing and correction methods that can be applied during permutations-based analysis of surface inputs (i.e., surface area, thickness, NPC of the two for volume, LGI, etc.).
- The FSL page on randomise
(https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise/UserGuide) suggests that cluster-based thresholding based upon the null distribution of the maximum cluster size/extent is passé in comparison to using the maximum cluster mass. PALM offers both extent and mass options, but does this same sentiment for functional data apply to SBM?
I don't know. You can run PALM on FS data. See http://freesurfer.net/fswiki/FsPalm
- My understanding is that LGI is already smoothed at ~ FWHM=15
during default processing of pial_lgi. If one were to look at area, thickness, and LGI, is there value to all variables being at the same level of smoothing (e.g., area @ 20mm, thickness @ 20mm, LGI @ 5mm)?
Probably not for LGI and thickness or area. I would guess not for any of them since they are telling you very different things that will have their own level of spatial correlation.
- Whereas earlier papers (e.g., #2) advocated for p < .005 and
cluster size > 10 voxels for functional analyses, for example, a number of folks have more recently suggested a high initial CFT (e.g., CFT=.001) and FWE = .05 (#s 3 & 4). Do any similar simulation results or suggestions exist for SBM rather than fMRI?
For permutation, the concerns about these thresholds to not apply -- you can use CFT=.05, which is nice. For random field correction, then the concerns do apply but they are not as bad as in fMRI (see #1)
- Does anyone know of guidance on FWE (or FDR or voxel/vertex-wise
for that matter) thresholds for TFCE-corrected results?
Would they be any different than other types of correction?
- Does anyone know of similar papers about FPRs in surface-based
analyses that /also/ consider false negative rates as well?
Not that I know of. You'd have to have a gold standard as to what effect you expect in order to determine the false neg rate.
- Have any journals made similar declarations about SBM/morphometry
results like NeuroImage: Clinical did for functional data (#5)?
Not that I know of.
Although I'm neither the first nor the last person to ask these questions, I wanted to gather my thoughts and hopefully tap into the collective wisdom of this listserv community. My apologies if this email is seen as incendiary or faux pas, but I'm new to this world and would love any and all suggestions.
Thanks in advance for sharing any thoughts and helping rectify my naiveté! Dan
1: Greve, D. N., & Fischl, B. (2018). False positive rates in surface-based anatomical analysis. /NeuroImage/, /171/, 6-14. 2: Lieberman, M. D., & Cunningham, W. A. (2009). Type I and Type II error concerns in fMRI research: re-balancing the scale. /Social Cognitive and Affective Neuroscience, 4/(4), 423-428. 3: Eklund, A., Nichols, T. E., & Knutsson, H. (2016). Cluster failure: why fMRI inferences for spatial extent have inflated false-positive rates. PNAS, 113(28), 7900-7905. https://doi.org/10.1073/pnas.1602413113 4: Woo CW, Krishnan A, Wager TD (2014) Cluster-extent based thresholding in fMRI analyses: Pitfalls and recommendations. /NeuroImage, 91/, 412–419. 5: Roiser, J. P., Linden, D. E., Gorno-Tempinin, M. L., Moran, R. J., Dickerson, B. C., & Grafton, S. T. (2016). Minimum statistical standards for submissions to Neuroimage: Clinical. /NeuroImage: Clinical/, /12/, 1045.
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Daniel Leopold -
Greve, Douglas N.,Ph.D.