Hi Andrew, To be clear, ICA-FIX may not improve task fMRI data when it is run separately on each tfMRI scan. MELODIC separates signal from noise better when there are tons of time points going into the ICA, which isn’t always the case with single tfMRI scans. Also, the confound regression will suffer from limited degrees of freedom when conducted on shorter single tfMRI scans, which could lead to reduced signal via overfitting noise predictors.
These are the concerns that led us to omit ICA-FIX from tfMRI preprocessing in the HCP. However, those concerns are reduced by combining / concatenating across multiple task fMRI runs (after appropriately removing the mean and linear trend from each run separately). We are discussing implementing this revised, multi-scan FIX pipeline for future projects, and potentially retrospectively for the original HCP data. To evaluate analyses on FIX vs. MPP tfMRI data, I think it’s important to conduct the comparison separately using a priori ROIs for signal and noise. --Greg ____________________________________________________________________ Greg Burgess, Ph.D. Staff Scientist, Human Connectome Project Washington University School of Medicine Department of Psychiatry Phone: 314-362-7864 Email: [email protected] > On Feb 20, 2017, at 10:51 AM, Glasser, Matthew <[email protected]> wrote: > > We have found that ICA+FIX does improve the HCP taskfMRI data. It reduces > false positives/false negatives in the beta maps and increases Z statistics. > For short tfMRI runs, I would recommend combining across runs so that ICA+FIX > can better separate signal and noise and so that the clean up regression is > better conditioned. > > Peace, > > Matt. > > From: <[email protected]> on behalf of "Poppe, Andrew" > <[email protected]> > Date: Monday, February 20, 2017 at 9:28 AM > To: "[email protected]" <[email protected]> > Subject: [HCP-Users] Comparing ICA-FIX results to results without ICA-FIX > > Hello, > > Our group is trying to decide if, and to what extent, using ICA-FIX improves > GLM results in our data analysis. We analyzed the same data with the same > models once using FIX prior to calculating stats and again without using FIX. > > We want to know how we might compare the two. > > One idea we had was to determine how well the model fit, as a reduction in > noise in the data should hopefully produce a more consistent model fit. > > To measure this, I tried to use the VARCOPEs associated with each COPE as a > measure of variability of the model fit. However, because the scale of these > images varied with the scale of the input data, I decided to divide each > VARCOPE image by the variance-across-time of the input data, producing > something like an R-squared image. Another method would be to calculate the > variance of the res4d image and divide that by the total variance and then > subtract from an image of all 1's. That should hopefully produce an R-squared > for the entire GLM model, as opposed to an individual COPE. > > I'm wondering if I'm fundamentally thinking about these images wrong, and > especially if there is a better way to compare the two analysis streams. > Without ground truth, is it possible to say whether FIX improved the analysis > in a GLM framework? > > Thanks, > > Andrew Poppe, PhD > Olin Center > Institute of Living > Hartford Hospital > > This e-mail message, including any attachments, is for the sole use of the > intended recipient(s) and may contain confidential and privileged > information. Any unauthorized review, use, disclosure, or distribution is > prohibited. If you are not the intended recipient, or an employee or agent > responsible for delivering the message to the intended recipient, please > contact the sender by reply e-mail and destroy all copies of the original > message, including any attachments. > _______________________________________________ > HCP-Users mailing list > [email protected] > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > > > > The materials in this message are private and may contain Protected > Healthcare Information or other information of a sensitive nature. If you are > not the intended recipient, be advised that any unauthorized use, disclosure, > copying or the taking of any action in reliance on the contents of this > information is strictly prohibited. If you have received this email in error, > please immediately notify the sender via telephone or return mail. > > _______________________________________________ > HCP-Users mailing list > [email protected] > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. _______________________________________________ HCP-Users mailing list [email protected] http://lists.humanconnectome.org/mailman/listinfo/hcp-users
