Based on the original paper of Shen, the confounding variables are only used in edge selection (AFAIK). I have implemented that in versions 0.0.5-0.0.6, but after second thought, I resorted to my method, i.e., using residuals after regressing all the confounds variables both for the connectivity matrices and behavioral measures.
This is open for discussion. However, for now I am satisfied with this "residual" method, although the predicted values are not so well-fitted for explanation.
Based on the original paper of Shen, the confounding variables are only used in edge selection (AFAIK). I have implemented that in versions 0.0.5-0.0.6, but after second thought, I resorted to my method, i.e., using residuals after regressing all the confounds variables both for the connectivity matrices and behavioral measures.
This is open for discussion. However, for now I am satisfied with this "residual" method, although the predicted values are not so well-fitted for explanation.