Dear experts,
I ran FLICA on a sample of 41 subjects, tried several options and have questions as follows. Each subject had a DTI (FA) map, an fsl VBM (GM) map, two freesurfer (thickness and area) map and two fMRI (resting state component; obtained by MELODIC and dual regression) maps. “num_components” was set to be 10;
1. “opts.maxits”. When this option was set to be 1000, nothing went wrong. But when it was set to be 10000, I got this error message:
“Warning: Non-finite values found in assertalmostequal!
> In assertalmostequal at 37
In flica_reorder at 68
Error using flica_reorder (line 120)
Assertion failed.”
Looks like I got a non-convergence. Any suggestions?
2. “opts.lambda_dims”. When this option was not touch (default setting), I got a overfitted model. Please see the attached png file. And every component was entirely (100% weight) explain by the VBM modality.
So I followed the suggestion on FLICA wiki and set opts.lambda_dims = '', to avoid per-subject noise estimation. Then the results looked much reasonable. But the values of pseudo-Z-statistic of freesurfer modality is very low (-0.112 ~ 0.019), even in a component where it had a 40% weight. Any suggestions?
Any comments or suggestions are much appreciated.
Kai Wang
Postdoc @ University of Colorado Boulder
|