Hi nltools experts,
I have collected a clinical dataset containing the patients’ resting-state fMRI and the clinical symptom score (evaluated by clinical doctors), and I aim to examine the shared pattern between the brain activity and the symptoms.
Following the tutorial about inter-subject similarity analysis, I have built the neural similarity matrix and the symptom similarity matrix. However, given that the age and head motion in clinical dataset have an impact to the result, I also want to regressor out the effect of these covariance. Do you have any advice about that?
Thanks in advance.
Apologies for the delay. One potential suggestion is to use distance regression, where you predict the brain inter-subject representational dissimilarity matrix (RDM) from a linear combination of other distance matrices. This would include your symptom RDM and also head motion and age. For single variables, euclidean distance usually makes the most sense. The effect of symptom severity on the brain pattern dissimilarity will be conditioned on age and head motion. If you wanted to adopt a more robust approach akin to a spearman correlation, than you would want to rank your distance values before running regression. This would work for both distance and similarity, though distance might require less transformations of the data.