RSA searchlight: high similarity values around edges of brain and CSF

[JasminStein]

Hi everyone,

I have used the rsatoolbox to compute a whole-brain searchlight RSA comparing neural RDMs with model RDMs to test for the encoding of object category (houses and faces). Threshholding the resulting similarity maps at the individual subject level, I have noticed that (in addition to seemingly reasonable peaks in visual and fusiform areas) there seem to be systematic strong peaks around the edges of the brain and around ventricles. Is that a common phenomenon when using searchlights?

Thanks a lot for your help!
Jasmin

[foldes.andrei]

@JasminStein

I have used the rsatoolbox to compute a whole-brain searchlight RSA comparing neural RDMs with model RDMs to test for the encoding of object category (houses and faces). Threshholding the resulting similarity maps at the individual subject level, I have noticed that (in addition to seemingly reasonable peaks in visual and fusiform areas) there seem to be systematic strong peaks around the edges of the brain and around ventricles. Is that a common phenomenon when using searchlights?

I think so! I think its good practice to 1) use a gray matter mask when doing whole-brain searchlight that excludes the CSF 2) specify a minimum voxel count/proportion in terms of radius. The latter should diminish “edge effects”. Those voxels shouldn’t survive group-level cluster correction in any case…