Reply-To: | | [log in to unmask][log in to unmask]>[log in to unmask]> >Date: 2010/12/1 >Subject: Re: [SPM] Is there the circularity of data in PPI? >To: <mailto:[log in to unmask]>[log in to unmask] > > >Don > >I've done both ways. I've used a group analysis >to define VOIs for PPI and I've also used >coordinates obtained from previous studies in >the literature to define VOIs for PPI analysis. >If the coordinates for the VOIs are in close >proximity to each other the results are generally similar. > >However, the issues of circularity are not just >limited to ROI selection for behavior/activation >correlations and multiple comparisons issues. It >really is a more fundamental issue -- basically >you are using the results from one analysis >(traditional subtraction functional >specialization analysis) to define the >parameters for a separate analysis (PPI analysis of functional integration). > >This is actually one of the issues that gets >consistently raised about seed based fcMRI >methods such as PPI versus multivariate methods >such as independent component analysis. "Seed >based fcMRI is constrained by the fact that >they rely strongly on a priori selection of >specific seed regions rather than allowing for >the characteristics of the network to identify >and locate the node regions" (paraphrased from >Tomasi and Volkow, PNAS, 127, 2010). That is a >round about way of saying they that seed-based methods are biased. > >Good discussion though. In my opinion, its best >to define the VOIs for PPI analysis based on >prior studies in the literature to avoid >potential issues with circularity and bias. > >Cheers > >Mehul > > >---------- >From: MCLAREN, Donald >[<mailto:[log in to unmask]>[log in to unmask]] >Sent: Wednesday, December 01, 2010 11:44 AM >To: Mehul Trivedi >Cc: <mailto:[log in to unmask]>[log in to unmask] > >Subject: Re: [SPM] Is there the circularity of data in PPI? > >The issues of those papers was that you can >extract a voxel or set of voxel to extract data >from and then repeat the analyses on the subset >of regions as that results in selection bias. >However, you can use the same data set and >correlate it with behavior and you will still >get the same result, if the relationship is >strong enough. The issues raised basically deal >with masking that reduce the MC problem. > >Now turning to PPI, previous posts have tended >to suggest a combination or anatomical and >functional definitions. For example, in region >X, pick all voxels in the subject that have >p<0.005, where region X can be defined from the >group functional map or an anatomical region. >I've cautioned against using the threshold >approach it biases the selection of voxels. In >that case, you might be argue that you are >double-dipping; although, clearly some paradigms >would necessitate this approach given the >spatial disparity across individuals (see >working memory tasks). That brings us back to >anatomy versus functionally defined regions. > >If your question is about the connectivity with >an anatomical region, then one should use the >anatomical region; however, anatomical regions >are often heterogeneous in function leading to a >poor result. Small anatomical regions should be fine. > >If your question is about the connectivity with >a functional region, then one should use the >group map. This is not double-dipping because >you are asking a fundamentally different >question. In this approach, you are not reducing >the MC problem, you are not ignoring the task as >the task is still in the model, and you are not >asking a question about the task evoked activity >(In my PPI approach -- see OHBM2008 abstract, >and soon to be submitted paper, each condition >is modeled separately and ALL conditions are >included - not a subset). The above referenced >papers deal with these three issues. > > >Best Regards, Donald McLaren > >2010/12/1 Mehul Trivedi ><<mailto:[log in to unmask]>[log in to unmask]> >Hi Feng > >I think that is exactly what Stephen means. If >you use a group analysis to define your VOI for >a PPI analysis using the same dataset, you are >guilty of double dipping. There have been >several recent pertinent discussions regarding >the topic of circularity in neuroimaging >research, specifically related to fMRI >activation / behavior correlations, but would >also be relevant to VOI definition as it relates to PPI analysis. > >My approach with PPI seed VOI selection has been >to obtain seed VOIs from multiple different >studies in the literature, which avoids issues with circularity. > >Hope this helps > >Mehul Trivedi > >References: > >1) Vul et al (2009) Puzzlingly High Correlations >in fMRI Studies of Emotion, Personality, and >Social Cognition, Perspectives on Psychological Science 2009 4: 274 > >NOTE: There are several commentaries associated >with this article that are worth reading as well. > >2) Poldrack and Mumford (2009) Independence in >ROI analysis: where is the voodoo? Social, >Cognitive, and Affective Neuroscience, 4, 208-213 > >3) Kriegeskorte et al (2009) Circular analysis >in systems neuroscience: the dangers of double >dipping. Nature Neuroscience 12, 535 - 540 (2009) > > >________________________________________ >From: SPM (Statistical Parametric Mapping) >[<mailto:[log in to unmask]>[log in to unmask]] >on behalf of Feng Lu >[<mailto:[log in to unmask]>[log in to unmask]] >Sent: Wednesday, December 01, 2010 7:01 AM >To: <mailto:SPM@JI4aÙ„¸MÀMÒ |