If you matrix multiply the component spatial maps by the component timeseries you can recover the original data minus the error term of the ICA (which is largely unstructured noise).
Matt.
Hello all,
I don't have an excellent mathematical background and I would like to understand better ICA. What is the meaning of the IC weights in the colour bar of Melodic? Are some voxels more “important”? On one hand I read that in an IC map all the voxels share the same timecourse, and on the other hand I read that the higher IC weights mean how much the timeseries of that voxel contributes to the IC signal, I'm confused. Is IC timeseries = weight_voxel1 * timecourse_voxel1 + weight_voxel2 * timecourse_voxel2 and so on?
Thank you,
Best wishes,
John
