 |
 |
Hello, Thanks for the melodic command line tip. I ran into this RAM limitation recently, so I increased the RAM from 2 GB to 8 GB. I am running FSL on a VM. I have a total of 16 GB RAM, and have allocated 8 GB to the VM. Also, on the FSL support forum and FAQ site I see the advice about RAM, but don't see specific RAM recommendations. If I allocate more, I think I may have an unstable situation, as I understand the support outside needs to be equivalent. Unless someone can advise me otherwise? Should this be enough to run Concat ICA on 15 subjects resampled to 2mm? I do see that the bubble help that comes up on Melodic giving a tip about this. It states: "for multi-subject Tensor ICA (although I am running concat), you probably need to set this to 3 or 4 mm so that the combined multi-subject data is not TOO large for processing to complete". Would resampling to 2mm interfere with Concat ICA? At the last run, the error occurred with PCA dimension estimation was called. Could this be "too big"? Or as we thought before, just a RAM problem. Reminding myself why I am resampling to 2 mm in the first place is because the stat images after dual regression had lots of activation outside the brain. So, we thought this was due to the standard image being at 2 mm and the stat images at 4 mm. So we upsampled the standard to 4mm. But still the activations were outside the brain. The bg_image showed the same. Like a bigger brain sitting on a small brain image. So I started scrutinizing the registration. And decided to increase the parameters from linear to warp between the functional and standard. And to avoid some error in matrix transform or whatever is going on with these stat images, put the resample to the same dimension as the standard inside of melodic. Things went nicely, the functional data from each subject was warped to the standard, the images normalized, but then at PCA it stopped. Most gratefully, Varina