Significant below chance values !

[CarolineLndl]

Hello Marseille-MVPA community  I have something wrong with my results but cannot found where it come from. Maybe I can found some help here. We run a whole brain searchlight MVPA analyse (the script is upload on this github, AudioTact_Project) on brain fMRI dataset. The experimental protocol included 3 Modalities (Haptic, Audio and AudioHaptic) and 2 textures (Rough Smooth). We run different leave2runs out analyses at individual level then [accuracy maps – 0.5] were entered in SnPM (5000 permutations, 0.05 FWE-cluster corrected) for group level analyses.

• Intramodal decoding: in each modality I trained the classifier to decode the texture. The results were very powerful, and I have observed common regions that decode the texture in each of the 3 intramodal decoding (A, H and AH).
• Cross-modal decoding: (I trained the classifier on one modality and tested on another one). I run the analysis and find significant below chance decoding in some regions (mostly the common regions of intramodal decodings). I was expected the exact opposite!
• I thought that perhaps the decoding relationship was strictly opposite in one modality compared to the other. To test this, I performed a univariate analysis that confirmed that the beta values in these regions increase in all modalities for the rough texture compared to the smooth texture. So, in the same direction
• Finally, I performed cross-modal decoding under bimodal and unimodal conditions (AH train and A or H test). I found significant, expected results in AH>A decoding but again significant reverse decoding (below chance) in the expected regions for AH>H decoding. I have the impression that something is wrong, I can't believe that the brain decodes in a totally opposite way in one modality compared to another and in the expected regions. The results of the univariate contrasts (Rough>Smooth) reinforce this feeling.

Have you ever had these kinds of results? Do you have any idea what I could test to better understand what is going on? Maybe plot some of the outputs of the decoding, but I don't know which ones can help me?

Merci pour votre aide ! Caroline

[CarolineLndl]

I have posted the issue if want to follow the update : https://neurostars.org/t/significant-below-chance-decoding-searchlight-cross-validation/19958/5

[JeanLucAnton]

In relation to this problem in the cross-modal decoding and as we discussed it by email, I was thinking :

• either there is a bug somewhere (like the inversion of R and S labels between the different modalities. Perhaps an inversion between H on the one hand and A and AH on the other?) but this is apparently unlikely.
• or the cerebral coding of the Rough and Smooth features is opposite between the A and H modalities. That is to say that a spatial pattern that would rather code Audio-Rough would rather code Haptic-Smooth (and symmetrically Audio-Smooth -> Haptic-Rough). Indeed, how can we affirm that, for such and such a brain region, what we have called Rough in Haptics generates the same response pattern as what we have defined as Rough in Audio? Note that I am talking about the spatial pattern of the BOLD response (what is analysed by the MVPA), not the average intensity of the response to which you refer in your post: "I thought that perhaps the decoding relationship was strictly opposite in one modality compared to the other. To test this, I performed a univariate analysis that confirmed that the beta values in these regions increase in all modalities for the rough texture compared to the smooth texture. So, in the same direction"