Dear Alfonso, dear Colleauges
During my experience with CONN a couple of problematic issues appeared and I would be grateful to hear your advice. I really apologize if the list is somewhat long...
[b]1. Denoising step[/b]: should the list of confounds, next to White Matter, CSF, realignement and scrubbing, include also separate effects of conditions (e.g. Effect of condition A pre-intervention, Effect of condition A post-intervention, Effect of condition B pre-intervention, etc.)? As I know, CONN puts them on the list sometimes by default, sometimes not, but they are to choose manually. Which consequences does it have to have them or not on the list as confound?
[b]2. Preprocessing and ART[/b]: I would like to investigate the impact of different ART parameters on results. If I am right, the preprocessing routine has no option "skip already processed steps". Can I somehow change the parameters alone based on the already preprocessed data (e.g. to run this step and smooth)? By the way, is there any golden standard for ART parameters? The quality of my data is good, however using 3SD and 1mm instead of previous liberal 9SD and 2mm made my results weaker, particularly influenced dramatically the MVPA (the only other difference was having effects of conditions on the confound list), what makes me a bit confused. The question is of course, whether the lower threshold reduced the artifacts or functionally meaningful effects.
[b]3. MVPA[/b]: Should I limit the dimensionality (default is 64) or put "inf" there? Should I keep the centrality on? I don't exactly understand which consequences it has. Moreover, I try to understand what is, so to speak obligatory and what optional, because this choice impacts the results: for creating seeds for post hoc seed-to-voxel analysis, do I need to load all the components at the same time (e.g. eye(8)) or am I allowed to look at each component separately? Are both parametric and non-parametric stats accepted and which height and cluster thresholds should be used there? In my data, the non-parametric stats result much more conservative.
[b]4. ICA: CONN and GIFT[/b]: next to the variety of resting state analysis performed in CONN, I want to run ICA using GIFT, because it still offers a few additional options for visualization and statistical sorting of components. I preprocessed and denoised data with CONN. As I understand, the swau files are the ones to use as input for GIFT, however, how can I consider all the fine denoising steps, CompCor, ART, etc.? Or maybe there is an elegant way to switch between CONN and GIFT allowing for the maximal use of GIFT features, such as MDL criteria, ICASSO and statistical sorting of components and making statistics in CONN (or something like this)?
[b]5. Matlab warning:[/b] Should I bother with the following Matlab warning (appearing after the first-level step)?
Warning: Returning NaN for out of range arguments
> In spm_Fcdf (line 108)
In conn_glm (line 152)
In conn_process (line 4151)
In conn_process (line 46)
In conn (line 5396)
In conn_contrasthelp>conn_contrasthelpcallback (line 86)
In conn_contrasthelp>@(varargin)conn_contrasthelpcallback(handle,contrasts,get
(gcbo,'value')-1) (line 64
Again, I apologize for the long list. I would be highly grateful for your advice.
Best, Lucas
During my experience with CONN a couple of problematic issues appeared and I would be grateful to hear your advice. I really apologize if the list is somewhat long...
[b]1. Denoising step[/b]: should the list of confounds, next to White Matter, CSF, realignement and scrubbing, include also separate effects of conditions (e.g. Effect of condition A pre-intervention, Effect of condition A post-intervention, Effect of condition B pre-intervention, etc.)? As I know, CONN puts them on the list sometimes by default, sometimes not, but they are to choose manually. Which consequences does it have to have them or not on the list as confound?
[b]2. Preprocessing and ART[/b]: I would like to investigate the impact of different ART parameters on results. If I am right, the preprocessing routine has no option "skip already processed steps". Can I somehow change the parameters alone based on the already preprocessed data (e.g. to run this step and smooth)? By the way, is there any golden standard for ART parameters? The quality of my data is good, however using 3SD and 1mm instead of previous liberal 9SD and 2mm made my results weaker, particularly influenced dramatically the MVPA (the only other difference was having effects of conditions on the confound list), what makes me a bit confused. The question is of course, whether the lower threshold reduced the artifacts or functionally meaningful effects.
[b]3. MVPA[/b]: Should I limit the dimensionality (default is 64) or put "inf" there? Should I keep the centrality on? I don't exactly understand which consequences it has. Moreover, I try to understand what is, so to speak obligatory and what optional, because this choice impacts the results: for creating seeds for post hoc seed-to-voxel analysis, do I need to load all the components at the same time (e.g. eye(8)) or am I allowed to look at each component separately? Are both parametric and non-parametric stats accepted and which height and cluster thresholds should be used there? In my data, the non-parametric stats result much more conservative.
[b]4. ICA: CONN and GIFT[/b]: next to the variety of resting state analysis performed in CONN, I want to run ICA using GIFT, because it still offers a few additional options for visualization and statistical sorting of components. I preprocessed and denoised data with CONN. As I understand, the swau files are the ones to use as input for GIFT, however, how can I consider all the fine denoising steps, CompCor, ART, etc.? Or maybe there is an elegant way to switch between CONN and GIFT allowing for the maximal use of GIFT features, such as MDL criteria, ICASSO and statistical sorting of components and making statistics in CONN (or something like this)?
[b]5. Matlab warning:[/b] Should I bother with the following Matlab warning (appearing after the first-level step)?
Warning: Returning NaN for out of range arguments
> In spm_Fcdf (line 108)
In conn_glm (line 152)
In conn_process (line 4151)
In conn_process (line 46)
In conn (line 5396)
In conn_contrasthelp>conn_contrasthelpcallback (line 86)
In conn_contrasthelp>@(varargin)conn_contrasthelpcallback(handle,contrasts,get
(gcbo,'value')-1) (line 64
Again, I apologize for the long list. I would be highly grateful for your advice.
Best, Lucas