Weird NEGATIVE correlation between insular and S1 in resting-state fMRI


I have a repeated measure dataset, where all subjects were scanned twice (i.e., pre & post) with an interval of 7 days between scans by fMRI. After analyzing by FSL standard pipeline, I got very weird group-level results using insular as ROI.

Pre: the functional connectivity between insular and S1 is Bilateral NEGATIVE!!! (with a large number of voxels)
Post: the functional connectivity between insular and S1 is positive (with a small number of voxels)
And paired t-test (Post > Pre) found large areas of difference in the sensorimotor region.

This result seems very strange to me as previous research has suggested that the test-retest reliability of resting-state functional connectivity is quite high. However, I am still unclear as to why the sensorimotor region appears to have a negative correlation during the pre-test.

Do any experts have any suggestions on this result? Are there any additional pre-processing methods such as denoising or motion correction that could improve the results?

Many Thanks!!!

Hi @iwannabebuer,

There are a lot of unknowns here that could provide some hints.

  1. how long are your scans (TR, number of volumes, etc)? Longer scans are more likely to converge for better reproducibility.

  2. would whatever happened pre-to-post influence resting state findings? Was there some kind of intervention?

  3. what kind of denoising did you use? Did you look at how parameters like motion varied between pre and post? Unmitigated motion artifacts will artificially increase connectivity (since noise tends to be correlated with itself).

  4. Have you looked at your insula seed to make sure the anatomical correspondence between pre and post looks okay?


Hi Steven,

Thank you so much for your reply, I appreciate you taking the time to think about it. Please see a more extended description of the issue below:

  1. Our rs-fMRI data were acquired using EPI sequence using an axial slice orientation (37 slices, TR/TE=2000/30 ms, 200 volumes) covering the whole brain.

  2. Thank you very much for pointing it out. Actually, during the pre-test, we needed the subjects to continue with additional tasks after resting-state scans. So we tied the stimulator to the subject’s right arm in advance, which will continuously give 32 degrees Celsius (assumed skin temperature). While in the post-test, since no additional tasks were required, we did not tie the stimulator to the arm. My colleagues and I have discussed this issue, and we do not think that only 32 °C (89.6 °F) of stimulation can cause such a significant change in functional connectivity.

  3. I used using various methods to denoise, including spatial and temporal filtering, motion correction, independent component analysis, despiking, and removing wm&csf signals. I also excluded all data with a head movement of more than 2.0 mm. Actually, I have tried different combinations of denoising methods and even used the original images for analysis, but the trend of the functional connectivity results did not change.

  4. yes. Native-space and standard-space masks all look fine to me.


Have you looked at insula connectivity to other regions to see if there are other changes pre-post? That is, are the trends you are observing global or limited to particular regions?

I would be concerned about what you raise in point 2, as I would imagine that having something tied to an arm would lead to some somatosensory signal that could confound things.