Back to the main page.

Bug 2935 - ft_artifact_zvalue does not detect jumps

Reported 2015-07-20 18:36:00 +0200
Modified 2016-01-14 21:27:37 +0100
Product: FieldTrip
Component: core
Version: unspecified
Hardware: PC
Operating System: Linux
Importance: P5 normal
Assigned to: Jan-Mathijs Schoffelen
Depends on:
See also:

Mats van Es - 2015-07-20 18:36:01 +0200

when using ft_artifact_zvalue to detect squid jumps (as used in the tutorial here jumps are not detected. I even tried artificially making a very large jump by changing the data, still it is not detected by the algorithm: data_short.trial{1}(132,1:700)=0; %channel 132 is MRC11 data_short.trial{1}(132,701:1441)=1; %data 'jumps' up to 1 and stays there (massively bigger jump than actual squid jump). I added a link to a few trials of data so you can try this yourself. Here, the first trial has the artificial jump in there, the second trial a 'real' jump. Download the data here: Note: I tried the algorithm on Linux using matlab2013a, matlab2014a and matlab2015a, all showing the same problem.

Jan-Mathijs Schoffelen - 2015-07-20 22:55:26 +0200

created /home/common/matlab/fieldtrip/data/test/bug2935.mat and checked. This particular data has a very strong cardiac artifact, causing the cumulated Z-value (across channels) to peak with higher values than the isolated jump. Setting cfg.artfctdef.zvalue.cumulative to 'no' does the trick here. @ Mats: try and understand the rationale for taking either a cumulation across channels, or alternatively select the maximum value across channels (the former looks for phenomena that are shared across several channels at the same time, the latter looks for channel isolated phenomena: although jumps typically occur in isolation, they are usually gigantic relative to the surrounding cumulative z-values, unless there are strong cumulated artifacts that are picked up with the same processing steps).could you please update the documentation on the wiki adding a small paragraph about the cumulative option?