Back to the main page.

Bug 2229 - Strange results of ft_sourcestatistics

Reported 2013-08-01 10:34:00 +0200
Modified 2015-07-15 13:31:20 +0200
Product: FieldTrip
Component: core
Version: unspecified
Hardware: PC
Operating System: Windows
Importance: P3 normal
Assigned to: Jim Herring
Depends on:
See also:

Kaoru Amano - 2013-08-01 10:34:39 +0200

I'm struggling with ft_sourcestatistics within subject. If I run ft_sourceanalysis without setting cfg.rawtrial = 'yes'; and calculate, say, NAI, the result looks fine. If I run ft_sourceanalysis with just adding cfg.rawtrial = 'yes';, and run ft_sourcestatistics between pre and post, the result looks completely different from NAI calculated without cfg.rawtrial = 'yes';. When I'm using the older version of FT, NAI and stat looked similar if I remember correctly... The attached file includes data and script for the comparison.

Kaoru Amano - 2013-08-01 10:55:19 +0200

Since the file was too big to attach, please download it from here.

Robert Oostenveld - 2013-08-01 13:53:14 +0200

I downloaded the data and copied it to /home/common/matlab/fieldtrip/data/test/bug2229 for internal use.

Jim Herring - 2013-09-24 16:55:46 +0200

Assigned myself

Jim Herring - 2014-01-29 15:11:30 +0100

Bug does not seem to be a problem anymore as output is highly similar (r=-0.99) in both cases. The correlation is only inversed now because in the test-script the difference between post and pre is plotted while the stat values are calculated on the difference between pre and post (A-B vs B-A).

Kaoru Amano - 2014-01-30 04:29:49 +0100

(In reply to Jim Herring from comment #4) Dear Jim Thank you so much for taking time to test the code. Now I updated my FT repository, and found that r=-0.99. Still, left-right of the estimated source seems to be flipped in both averaged and single-trial cases. Iso-contour map suggests strong activation in the left hMT+, but the estimated source is at around the right hMT+. When I run the test code with older FT (13/07/30 version), activity in the left hMT+ was estimated correctly (with LCMV averaged source analysis not with LCMV single source analysis). Can you imagine what's the problem? kaoru