 |
 |
Hi Shruti, On 20 Feb 2017, at 14:29, Shruti Narasimham <[log in to unmask]<mailto:[log in to unmask]>> wrote: Dear Janine, Thank you so much for this explanation, but I'm a still bit confused due to the following : You said the dual regression and randomise results are a within-network analysis. But as per what's written in this link: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/DualRegression/Faq#What_does_it_mean_if_I_find_a_dual-regression_result_outside_a_given_RSN.3F , if an activation lies outside a particular RSN for the corresponding IC, it can mean that the "connectivity" of this area with the main regions of this RSN is different in the two groups”. Yes, that’s right. Dual regression is a voxel-wise analysis that looks at whole-brain correlation with the corresponding group IC that was fed into the analysis. It might reveal that some brain areas outside of the corresponding IC are part of this network in patients but not in controls (or vice versa). However, you can essentially still think of this as a within-network analysis, because it looks at changes in the shape of a single network (for example of the DMN), and not at connectivity between two networks (for example the DMN - DAN). I am confused then what this connection difference (between region A and B) means w.r.t the connection (between region / IC A and B) in the FSLNets between node analysis. Isn't it the same between region analysis and results in both the cases then? Dual regression is a spatial whole-brain analysis which identifies brain regions that have increased or decreased connectivity with IC A. FSLnets is a temporal analysis which keeps the spatial maps fixed and looks at correlations in timeseries between IC A and B. For example, if you feed a set of maps including the DMN and the DAN into a dual regression analysis, then the results could show that some specific region is part of the DMN in patients but not in controls. However, this doesn’t tell you anything about the connectivity between the DMN and the DAN (which is exactly what an FSLnets analysis would tell you exactly this). Cheers, Janine Regards, Shruti On 20 February 2017 at 13:03, Janine Bijsterbosch <[log in to unmask]<mailto:[log in to unmask]>> wrote: Hi Shruti, There are a number of complex differences between dual regression and FSLnets which could lead to these findings, and it’s very hard to be confident about what is driving this. Some of the most important difference are probably the following: Firstly, dual regression is primarily a within-network analysis which looks for brain areas that differ in their connectivity with the relevant group-level network between patients and controls. FSLnets, on the other hand, is mainly a between-node analysis which investigates changes in correlation between nodes. If you are using the same regions for both the dual regression and the FSLnets analysis (which may in itself not be ideal), then it is possible that there is a difference in connectivity strength between two regions, but that connectivity within each network doesn’t change. Secondly, the statistical sensitivity may differ between the two approaches due to differences in multiple comparison correction. For example, if you are working with a small number of regions then FSLnets may be more sensitive to detect changes than dual regression, because you are correcting across fewer edges. In general, I would be very careful to draw any inferences from these results, because it is not possible to directly compare them, and there are many influences that may play a role. Cheers, Janine On 20 Feb 2017, at 11:48, Shruti Narasimham <[log in to unmask]<mailto:[log in to unmask]>> wrote: Hello, What is the difference in running an FSLNets analysis vs. TFCE randomization with dual regression analysis? I understand that their working and output is very different, but I was wondering what could contribute to a difference in the results obtained in these two cases. For example, if there is no significant difference in FC between patients and controls between 2 regions, but there is a significant difference in correlation / edges between these 2 regions obtained from the FSLNets analysis, what could be the correct inference for such a situation? I want to be sure before drawing any inferences, therefore request for help / any information on this. Thank you. Regards, Shruti ----- Dr Janine Bijsterbosch Postdoctoral Researcher FMRIB Centre, University of Oxford John Radcliffe Hospital Oxford, United Kingdom [log in to unmask]<mailto:[log in to unmask]> ----- Dr Janine Bijsterbosch Postdoctoral Researcher FMRIB Centre, University of Oxford John Radcliffe Hospital Oxford, United Kingdom [log in to unmask]<mailto:[log in to unmask]>