Dear Jukka, Samu, and colleagues at Elekta and elsewhere,

I found a bug in  MaxFilter that produces replacement of data chunks with copies of a preceding data chunk in sequence repetitively.

In other words, a chunk of data is repeated several times thereby replacing some of the data that should follow the data chunk that is repeated.

I attach a screenshot showing an example data set (with the problem, above; and fixed, below).

The trigger channel and other channels are not repeated, just the MEG channels are repeated. Needless to say, this will result in problems for subsequent ERF analysis.

The bug is produced when the hpie and hpig parameters are active because the error and/or goodness-of-fit limits are passed.

This bug happens whether you are just doing head estimation (estimating the quaternion parameters) or actually doing movement compensation.

If I set the hpie parameter to a large number and the hpig parameter to zero, the data replacement bug does not happen anymore.

The size of the data chunks seems corresponds to the tSSS time window size specified with -st (i.e., if st is 4 then the chunk of data that is repeated is 4 sec long).

The MaxFilter version I was using when I found this was 2.2.15, but I checked with version 2.2.10 and the same (or similar) bug is present related again to hpie and hpig (producing the same exact data replacement with repetitive chunks).

Again the solution is to set hpie to a large number (e.g., 1000000) and hpig to 0.

Please make a note, and let others know as this could have been affecting the data sets of many users world-wide who have been performing head estimation or head movement compensation.

Best regards,

Rey


--
Rey R. Ramírez, Ph.D.
MEG Brain Imaging Center
Institute for Learning & Brain Sciences
Portage Bay Building
Box 357988
University of Washington
Seattle, WA 98195-7988
e-mail: [log in to unmask]
Tel: (206) 543-6484
Fax: (206) 221-6472
Cell: (619) 709-3590


Inline image 2