Indipendent Component Analysis: MULTICOMBI quality check

Submitted by Giuseppe on
Hi, is there anyone who knows the ICA code MULTICOMBI of Prof. Petr Tichavsky? I am using it, but I am not sure I am able to evaluate correctly the accuracy of estimations.

Thanks for your answer. I think to use the estimate ISR matrix (Interference to Signal Ratio) to estimate the accuracy. I can write it in case, but the point is that: because the program automatically order the output signals from the most important to the less important, I expect to find that sums of the elements of the rows increase from the first row to the last. It does not happen, there is no order apparently. It may be not a peculiarity of MULTICOMBI code. Do you have any idea? Best, Giuseppe

Hi Giuseppe, Sounds its not a usually used ICA algorithm in fMRI (e.g., Infomax, FastICA). I am not sure if I get you correctly, since you would like to assess the ICA decomposition's accuracy, usually in fMRI community, it should be done by means of goodness-of-fit coefficient, calculated by comparing IC (spatial or temporal pattern) with the golden standard (a mask derived from brain-region-label or a task-related time course). I don't know about ISR matrix, maybe you can refer to the author of COMBI code? As for the order of the ICA result, I think it should be in no order, unless you define a goodness-of-fit by fitting ICA result to golden standard and then order those ICs based on GOF. Sorry for can't helping you.... Han
Forums