Thanks Tom, I was checking what the COBIDAS report was mentioning,
hoping that it would provide evidence to editors that they are asking
for the impossible:
http://www.humanbrainmapping.org/files/2016/COBIDASreport.pdf#[{%22num%22%3A58%2C%22gen%22%3A0}%2C{%22name%22%3A%22XYZ%22}%2C72%2C287.25%2C0]
The relevant sentences are:
"If cluster-wise inference is used, the cluster statistic (e.g. size,
mass, etc) should be included."
"Whenever possible, provide effects sizes at the selected coordinates
together with 95% confidence intervals."
Not entirely explicit but hopefully enough.
Best regards,
Guillaume.
On 19/02/2021 17:46, Thomas Nichols wrote:
> Hi Jeff,
>
> The lack of confidence intervals for cluster size is unfortunately an
> established limitation of random field theory and most standard
> methods... in fact that's a bullet point on one my SPM talk slides!
>
> This limitation motivated the work that Guillaume referenced, the
> Bowring et al work; there are now two papers, one for sample mean, one
> for Cohen's d:
>
> https://pubmed.ncbi.nlm.nih.gov/31533067/
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpubmed.ncbi.nlm.nih.gov%2F31533067%2F&data=04%7C01%7C%7C7acd7d7a31654d75956c08d8d4fe5183%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C1%7C637493536046639702%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=1D5jVmD%2BqfBegwPZIuVbfRtfVHxS71h6xm0oDVIsrV8%3D&reserved=0>
> https://pubmed.ncbi.nlm.nih.gov/33166643/
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpubmed.ncbi.nlm.nih.gov%2F33166643%2F&data=04%7C01%7C%7C7acd7d7a31654d75956c08d8d4fe5183%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C1%7C637493536046639702%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=UKo57dJVIXlyApAs8lvTDYnzGipT4FEu46ZmXYmYEuU%3D&reserved=0>
>
>
> (One thing to appreciate is that a confidence interval is always on a
> parameter that is meaningful in the population, e.g. like the mean or a
> proportion... you never get CI's for a T-value, right? Well, likewise,
> you can't talk about a spatial analog of a cluster defined on a test
> statistic image, since a test statistic is not an estimate of some
> feature of a population, but just a device to conduct a hypothesis
> test. Fortunately, for many models T images are proportional to Cohen's
> d, and so any clusters defined on a T-image with some t-valued
> cluster-forming threshold can be equivalently found in a Cohen's d image
> for some Cohen's d threshold).
>
> The code for the Cohen'd d work is
> https://github.com/AlexBowring/Confidence_Sets_Manuscript but we haven't
> had a chance to make it really user friendly or general.
>
> In short, while I'd love for you to try our method and let us know how
> it goes, for your paper I'd respond to the reviewer that cluster size is
> well-known to lack confidence intervals with standard methods.
>
> -Tom
>
> On Tue, Feb 9, 2021 at 6:29 PM Guillaume Flandin <[log in to unmask]
> <mailto:[log in to unmask]>> wrote:
>
> Hi Jeff,
>
> Now is time to phone a friend. Tom, how do you appease editors here?
>
> Best regards,
> Guillaume.
>
>
> On 09/02/2021 17:55, Jeff Browndyke wrote:
> > Hi, Guillaume.
> >
> > We ran a standard GLM with a peak voxel threshold of p<0.001 and
> cluster threshold of pFWE<0.01 in CONN looking for group differences
> in one of the whole brain voxel-wise functional connectivity
> metrics. The journal is is requesting CI for any stated p-values,
> so in this case I took it to mean the CI for p-values reported for
> cluster extent significance (e.g., cluster A extent k=459,
> p-FWE=0.003).
> >
> > Warm regards,
> > Jeff
> >
> >
> >> On Feb 9, 2021, at 12:26 PM, Guillaume Flandin
> <[log in to unmask] <mailto:[log in to unmask]>> wrote:
> >>
> >> Dear Jeff,
> >>
> >> It seems that for the script you are mentioning, one could nowadays
> >> directly call spm_graph.m. Concerning confidence intervals for
> >> cluster-level inference, are you looking for CI on the cluster
> extent?
> >> If so, it might not be easy, but see e.g. this:
> >>
> https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.sciencedirect.com%2Fscience%2Farticle%2Fpii%2FS1053811919307785&data=04%7C01%7C%7C8c0049706e0e4903b55108d8cd23dc46%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637484901195944274%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=E1rfGwfXALm2YZo0wjvRLMi263FH0jZ1WKNingaso%2F8%3D&reserved=0
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.sciencedirect.com%2Fscience%2Farticle%2Fpii%2FS1053811919307785&data=04%7C01%7C%7C7acd7d7a31654d75956c08d8d4fe5183%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C1%7C637493536046649697%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=rCcF0FFvHmJDZ74FeAkKVrU%2Bk6kA%2FiNMvJD0peU9RxI%3D&reserved=0>
> >> When you say "thresholded @ p-FWE<0.01", what was the cluster-forming
> >> threshold?
> >>
> >> Best regards,
> >> Guillaume.
> >>
> >>
> >> On 09/02/2021 13:38, Jeffrey Browndyke wrote:
> >>>
> >>> SPMers,
> >>>
> >>> I have used the following scripts in the past to derive std err and
> >>> confidence intervals for peak voxel results:
> >>>
> >>>
> https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fimaging.mrc-cbu.cam.ac.uk%2Fimaging%2FExtracting_beta%252C_standard_error_and_confidence_interval_for_a_coordinate&data=04%7C01%7C%7C8c0049706e0e4903b55108d8cd23dc46%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637484901195944274%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=WEym1XB3tW3D5DlvaRSeohBgNrSHR5VnU0Fc31%2BP6K8%3D&reserved=0
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fimaging.mrc-cbu.cam.ac.uk%2Fimaging%2FExtracting_beta%252C_standard_error_and_confidence_interval_for_a_coordinate&data=04%7C01%7C%7C7acd7d7a31654d75956c08d8d4fe5183%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C1%7C637493536046649697%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=QkhNaY5TqqAJ4Pypi0RWeJPLs5dVUeglNX9Ovs1271g%3D&reserved=0>
> >>>
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fimaging.mrc-cbu.cam.ac.uk%2Fimaging%2FExtracting_beta%252C_standard_error_and_confidence_interval_for_a_coordinate&data=04%7C01%7C%7C8c0049706e0e4903b55108d8cd23dc46%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637484901195944274%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=WEym1XB3tW3D5DlvaRSeohBgNrSHR5VnU0Fc31%2BP6K8%3D&reserved=0
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fimaging.mrc-cbu.cam.ac.uk%2Fimaging%2FExtracting_beta%252C_standard_error_and_confidence_interval_for_a_coordinate&data=04%7C01%7C%7C7acd7d7a31654d75956c08d8d4fe5183%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C1%7C637493536046659690%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=Xsa72jxFAYLReTG6ckw3YoDV4zYTtq7HOphNbIwWpEg%3D&reserved=0>>
> >>>
> >>> But, we are being asked to provide confidence intervals for the
> >>> cluster-level significance for blobs in one of our analyses
> (thresholded
> >>> @ p-FWE<0.01).
> >>>
> >>> Any pointers or guidance on where to find or how to derive this
> info?
> >>>
> >>> Thanks,
> >>> Jeff Browndyke
> >>>
> >>> Sent from my iPhone
> >>>>
> >>
> >> --
> >> Guillaume Flandin, PhD
> >> Wellcome Centre for Human Neuroimaging
> >> UCL Queen Square Institute of Neurology
> >> London WC1N 3BG
> >
>
> --
> Guillaume Flandin, PhD
> Wellcome Centre for Human Neuroimaging
> UCL Queen Square Institute of Neurology
> London WC1N 3BG
>
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|