help > RE: CONN connectivity results are not reproducing
Apr 16, 2018  12:04 PM | Alfonso Nieto-Castanon - McGovern Institute for Brain Research. MIT
RE: CONN connectivity results are not reproducing
Dear Ekaterina,

Thanks for doing this work! A couple of quick checks, just to make sure there are no other potentially confounding differences between the different machines SPM versions:

1) could you please check the SPM12 release numbers in the different machines? (you can check the spm release number using the syntax "spm version", I copied below the different SPM12 release numbers just for reference)
Date Update Revision Number
01-Oct-2014 6225
03-Jun-2015 6470
15-Jan-2016 6685
20-Oct-2016 6906
16-Nov-2017 7219

and 2) could you please run on the different machines spm_get_defaults('realign.estimate') as well as spm_get_defaults('unwarp.estimate') ? I see a small but noticeable difference in the estimated motion parameters between the first 2 and last 3 machines and I am trying to figure out why that might be the case. 

Thanks again!

Originally posted by Ekaterina Kondrateva:
Dear Alfonso,

My apologies for such a long response, I was retesting the experiment that was described previously on different machines.

Thus to conclude:

1. the results are 100% reproducible on one machine in time, it was just my bad: I've used *.nii structural files stacked in different software (SPM and Nibabel for the 3D image retrieval from DICOM). And it leads to some bias, that I've referred to CONN processing.

2. the results are not reproducing on different machines, indeed; and the bias is related to the processor core. On our test, the maximum Mean Average Percentage Error (MAPE) of the functional connectivity graph analysis results reached 8%.

Attaching the archive with 5 *.mat files (4 subjects in each). These files were processed with default settings on 5 different machines and should produce identical results: 1. Core i5, Win10 2. Core i5, macOS 3. Pentium, Linux, 4. and 5. Core i7, Win 10.

Could you please have a brief look if you can spot the differences?

Thanks so much!

Warm regards, Kate
Threaded View
RE: CONN connectivity results are not reproducing
Apr 16, 2018