help
486 Subscribers 
help > Second level analysis error
Showing 1-3 of 3 posts
Apr 3, 2017 12:04 PM | Larissa
Second level analysis error
ERROR DESCRIPTION:
Dear Alfonso,
I have 2 questions:
1. When running a seed-to-voxel analysis comparing 2 resting-state conditions in 10 subjects I get the following error message when selecting 'Results for all sources.' Note I'm doing an exploratory analysis: my between subjects contrast is 1 (for All subjects), Between conditions contrast is [1 -1] when comparing the 2 conditions.
Index exceeds matrix dimensions.
Error in conn_process (line 3937)
nrepeated=size(SPMall(n1).xY.VY,2);
Error in conn_process (line 51)
case 'results_voxel', [varargout{1:nargout}]=conn_process(16,varargin{:});
Error in conn (line 6681)
conn_process('results_voxel','doall','seed-to-voxel');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM12 + DEM FieldMap MEEGtools
Matlab v.2012a
storage: 1530.9Gb available
2. If I want to look at a specific network as an a priori seed/source for the seed-to-voxel design (e.g. the visual cortex), there are 4 network ROIs that CONN provides (ie. network.Visual.Primary/Ventral/Dorsal (L)/Dorsal (R). If I select all 4 visual networks as the seed ROI, I get a significant cluster as a result when selecting Results explorer. However, when I select each individual network ROI on its own, I do not find any significant clusters. My understanding is that my ROI is larger and is therefore more prone to find significance when selecting all 4 ROIs that contribute to the network in question. Does CONN do a multiple comparisons analysis with FDR if this is the case? Is it normally justified to analyze data as such, since the 4 ROIs contribute to the same network?
Thank you,
Larissa
Dear Alfonso,
I have 2 questions:
1. When running a seed-to-voxel analysis comparing 2 resting-state conditions in 10 subjects I get the following error message when selecting 'Results for all sources.' Note I'm doing an exploratory analysis: my between subjects contrast is 1 (for All subjects), Between conditions contrast is [1 -1] when comparing the 2 conditions.
Index exceeds matrix dimensions.
Error in conn_process (line 3937)
nrepeated=size(SPMall(n1).xY.VY,2);
Error in conn_process (line 51)
case 'results_voxel', [varargout{1:nargout}]=conn_process(16,varargin{:});
Error in conn (line 6681)
conn_process('results_voxel','doall','seed-to-voxel');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM12 + DEM FieldMap MEEGtools
Matlab v.2012a
storage: 1530.9Gb available
2. If I want to look at a specific network as an a priori seed/source for the seed-to-voxel design (e.g. the visual cortex), there are 4 network ROIs that CONN provides (ie. network.Visual.Primary/Ventral/Dorsal (L)/Dorsal (R). If I select all 4 visual networks as the seed ROI, I get a significant cluster as a result when selecting Results explorer. However, when I select each individual network ROI on its own, I do not find any significant clusters. My understanding is that my ROI is larger and is therefore more prone to find significance when selecting all 4 ROIs that contribute to the network in question. Does CONN do a multiple comparisons analysis with FDR if this is the case? Is it normally justified to analyze data as such, since the 4 ROIs contribute to the same network?
Thank you,
Larissa
Apr 3, 2017 06:04 PM | Alfonso Nieto-Castanon - Boston University
RE: Second level analysis error
Dear Larissa,
Regarding (1) thanks for reporting this bug, please try the attached patch and let me know if you still run into any issues (this patch is for release 17c, simply copy this file to the conn distribution folder overwriting the file with the same name there)
Regarding (2) when selecting all four seeds CONN is performing a multivariate test (instead of a univariate test for a single seed) looking at any effect across the four seeds. The multivariate test is going to automatically correct for the "multiple-test" scenario of having four seeds. It is mathematically possible but in practice somewhat unlikely that you will get a significant multivariate effect in the absence of any univariate effect. Could you confirm if, when looking at the univariate effects (separately for each seed), you are selecting a "two-sided" test? (by default the results explorer window for a univariate test comes up with the one-sided test -positive directionality- just to follow SPM convention, but in this case I am thinking that perhaps you may not be seeing any univariate effect because they may be negative / in the opposite directionality?)
Hope this helps
Alfonso
Originally posted by Larissa :
Regarding (1) thanks for reporting this bug, please try the attached patch and let me know if you still run into any issues (this patch is for release 17c, simply copy this file to the conn distribution folder overwriting the file with the same name there)
Regarding (2) when selecting all four seeds CONN is performing a multivariate test (instead of a univariate test for a single seed) looking at any effect across the four seeds. The multivariate test is going to automatically correct for the "multiple-test" scenario of having four seeds. It is mathematically possible but in practice somewhat unlikely that you will get a significant multivariate effect in the absence of any univariate effect. Could you confirm if, when looking at the univariate effects (separately for each seed), you are selecting a "two-sided" test? (by default the results explorer window for a univariate test comes up with the one-sided test -positive directionality- just to follow SPM convention, but in this case I am thinking that perhaps you may not be seeing any univariate effect because they may be negative / in the opposite directionality?)
Hope this helps
Alfonso
Originally posted by Larissa :
ERROR DESCRIPTION:
Dear Alfonso,
I have 2 questions:
1. When running a seed-to-voxel analysis comparing 2 resting-state conditions in 10 subjects I get the following error message when selecting 'Results for all sources.' Note I'm doing an exploratory analysis: my between subjects contrast is 1 (for All subjects), Between conditions contrast is [1 -1] when comparing the 2 conditions.
Index exceeds matrix dimensions.
Error in conn_process (line 3937)
nrepeated=size(SPMall(n1).xY.VY,2);
Error in conn_process (line 51)
case 'results_voxel', [varargout{1:nargout}]=conn_process(16,varargin{:});
Error in conn (line 6681)
conn_process('results_voxel','doall','seed-to-voxel');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM12 + DEM FieldMap MEEGtools
Matlab v.2012a
storage: 1530.9Gb available
2. If I want to look at a specific network as an a priori seed/source for the seed-to-voxel design (e.g. the visual cortex), there are 4 network ROIs that CONN provides (ie. network.Visual.Primary/Ventral/Dorsal (L)/Dorsal (R). If I select all 4 visual networks as the seed ROI, I get a significant cluster as a result when selecting Results explorer. However, when I select each individual network ROI on its own, I do not find any significant clusters. My understanding is that my ROI is larger and is therefore more prone to find significance when selecting all 4 ROIs that contribute to the network in question. Does CONN do a multiple comparisons analysis with FDR if this is the case? Is it normally justified to analyze data as such, since the 4 ROIs contribute to the same network?
Thank you,
Larissa
Dear Alfonso,
I have 2 questions:
1. When running a seed-to-voxel analysis comparing 2 resting-state conditions in 10 subjects I get the following error message when selecting 'Results for all sources.' Note I'm doing an exploratory analysis: my between subjects contrast is 1 (for All subjects), Between conditions contrast is [1 -1] when comparing the 2 conditions.
Index exceeds matrix dimensions.
Error in conn_process (line 3937)
nrepeated=size(SPMall(n1).xY.VY,2);
Error in conn_process (line 51)
case 'results_voxel', [varargout{1:nargout}]=conn_process(16,varargin{:});
Error in conn (line 6681)
conn_process('results_voxel','doall','seed-to-voxel');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM12 + DEM FieldMap MEEGtools
Matlab v.2012a
storage: 1530.9Gb available
2. If I want to look at a specific network as an a priori seed/source for the seed-to-voxel design (e.g. the visual cortex), there are 4 network ROIs that CONN provides (ie. network.Visual.Primary/Ventral/Dorsal (L)/Dorsal (R). If I select all 4 visual networks as the seed ROI, I get a significant cluster as a result when selecting Results explorer. However, when I select each individual network ROI on its own, I do not find any significant clusters. My understanding is that my ROI is larger and is therefore more prone to find significance when selecting all 4 ROIs that contribute to the network in question. Does CONN do a multiple comparisons analysis with FDR if this is the case? Is it normally justified to analyze data as such, since the 4 ROIs contribute to the same network?
Thank you,
Larissa
Apr 15, 2017 05:04 PM | Larissa
RE: Second level analysis error
Hi Alfonso,
Regarding 1), I still get an error when running a seed-to-voxel 2nd level analysis- when selecting 'Results for all sources" when overwriting the conn_process.m file you provided. In fact, any analysis was creating error messages. I had to revert back to the original conn_process.m file to have conn working. Note, I am using conn version 17a. You mentioned you compiled the patch using 17c- if that's causing the error, could I get the 17a patch?
ERROR DESCRIPTION:
Cell contents reference from a non-cell array object.
Error in conn_process (line 3798)
if isdir(fullfile(filepathresults2,foldername_back{nfolderbak})), foldername=foldername_back{nfolderbak}; break; end % backwards-compatibility with existing results
Error in conn_process (line 52)
case 'results_voxel', [varargout{1:nargout}]=conn_process(16,varargin{:});
Error in conn (line 6681)
conn_process('results_voxel','doall','seed-to-voxel');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM12 + DEM FieldMap MEEGtools
Matlab v.2012a
storage: 1497.5Gb available
Regarding 2): Yes, the solution was to select the two-sided test when looking for univariate effects.
Thanks,
Larissa
Regarding 1), I still get an error when running a seed-to-voxel 2nd level analysis- when selecting 'Results for all sources" when overwriting the conn_process.m file you provided. In fact, any analysis was creating error messages. I had to revert back to the original conn_process.m file to have conn working. Note, I am using conn version 17a. You mentioned you compiled the patch using 17c- if that's causing the error, could I get the 17a patch?
ERROR DESCRIPTION:
Cell contents reference from a non-cell array object.
Error in conn_process (line 3798)
if isdir(fullfile(filepathresults2,foldername_back{nfolderbak})), foldername=foldername_back{nfolderbak}; break; end % backwards-compatibility with existing results
Error in conn_process (line 52)
case 'results_voxel', [varargout{1:nargout}]=conn_process(16,varargin{:});
Error in conn (line 6681)
conn_process('results_voxel','doall','seed-to-voxel');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM12 + DEM FieldMap MEEGtools
Matlab v.2012a
storage: 1497.5Gb available
Regarding 2): Yes, the solution was to select the two-sided test when looking for univariate effects.
Thanks,
Larissa