 |
 |
Jeff, SnPM, or any nonparametric permutation method, depends on the data studied, so there's no equivalent way you could get results for just based on FWHM and search volume (even *if* you knew the pseudo t's effective DF). If you have a particular SnPM result, though MaxT(:,1) is the voxel-wise max permutation distribution, and the 95%ile of that will give you the corrected voxel-wise thresold, and the STCS structure's MxK(:,1) gives you maximume cluster size distribution. For more details see snpm_cp.m and snpm_STcalc.m -Tom On 2/19/08, Jeff Browndyke, Ph.D. <[log in to unmask]> wrote: > > Tom, > > This is very helpful, as usual. Thanks for taking the time to respond so > thoroughly. Most appreciated! > > One follow-up though. What if the results were SnPM, for which error DF > is not known because the analysis it based upon pseudo-t ? > > Cheers, > Jeff > > ------------------------------ > *From:* [log in to unmask] [mailto:[log in to unmask]] *On Behalf Of *Thomas > Nichols > *Sent:* Monday, February 18, 2008 8:50 AM > *To:* [log in to unmask] > *Cc:* [log in to unmask] > *Subject:* Re: [SPM] Calculating adjusted threshold from extent and > p-value data? > > Jeff, > > On Feb 11, 2008 5:08 PM, Jeff Browndyke <[log in to unmask]> wrote: > > > Is there a way to calculate an adjusted statistical threshold for an > > analysis outside of SPM given a stated p-value threshold (i.e., p < .01, > > uncorrected), extent threshold [k > 20 (40mm3)], and known voxel number for > > the analysis? > > > SPM uses the smoothness of the noise and the topology of the search region > to determine corrected P-values. Hence to duplicate the inferences outside > of SPM you'd have to have the complete RESEL count vector available, as > stored in SPM.xVol.R in SPM.mat > > If approximate inference is sufficient, and you know at least the FHWM of > the noise smoothness, the snippet below will give you some idea of the > corrected P-value for a given FWHM, assuming a spherical search region of a > given radius. More accurate P-values will be had by reading the help on > spm_resels and specifying a more accurate search region, and, most > accurately, using the RESEL vector from SPM.xVol.R of the relevant SPM > results. > > -Tom > > %%% See spm_resels for more > FWHM = [5 5 5]; % *Voxel* FWHM smoothness --- CHANGE FOR YOUR DATA! > L = 25; % Region size, voxels --- CHANGE FOR YOUR DATA! > SPACE = 'S'; % Type of region (here 'S'phere) > > > %%% See spm_P_RF for more > erdf = 50; % Error DF --- CHANGE FOR YOUR DATA! > STAT = 'T'; > df = [1 erdf]; > R = spm_resels(FWHM,L,SPACE); > v2r = 1/prod(FWHM); % Voxels-to-resels conversion > > > % P-value for particular cluster size, cluster forming threshold > ualpha = 0.01; % Uncorrected cluster-forming alpha > u = spm_invNcdf(1-ualpha); % Uncorrected cluster-forming threshold > > k = 20; % Clust size in voxels --- CHANGE FOR > YOUR DATA! > > % Corrected P-value > P = spm_P_RF(1,k*v2r,u,df,STAT,R,1) > > > > PS: For a cluster threshold result instead of P-value, see > http://www.sph.umich.edu/~nichols/JohnsGems5.html#Gem6<http://www.sph.umich.edu/%7Enichols/JohnsGems5.html#Gem6>... though that script currently requires a completed SPM analysis. A > little hacking and that script could allow you to specify just FWHM as > above. > > ____________________________________________ > Thomas Nichols, PhD > Director, Modelling & Genetics > GlaxoSmithKline Clinical Imaging Centre > > Senior Research Fellow > Oxford University FMRIB Centre > -- ____________________________________________ Thomas Nichols, PhD Director, Modelling & Genetics GlaxoSmithKline Clinical Imaging Centre Senior Research Fellow Oxford University FMRIB Centre