lowpass filtering data in sensor or source space

✉️ Mailing List Archive (read-only)
1

Hello,

I have used mne_compute_raw_inverse to produce the inverse solution for
raw data, and loaded this data into matlab.

However, I have inspected the data and would like to lowpass filter the
data further before extracting the statistics that I am interested in.

It seems like my options are to either low pass filter the raw sensor
data further before computing the inverse solution, or to filter the
source data directly in matlab. Is there any methodological reason to
prefer one method over the other?

If filtering the source data in matlab is preferred (or if both methods
are fine), could anyone recommend a particular filter type? I've looked
into documentation for filtering in matlab and it seems like there are a
tremendous amount of options.

Thanks,

2

Filtering in source space is very costly, number of sources >> number of
sensors, best filter to use which granted zero phase shift in Matlab is
eegfilt (attached).

Sheraz

Hello,

I have used mne_compute_raw_inverse to produce the inverse solution for
raw data, and loaded this data into matlab.

However, I have inspected the data and would like to lowpass filter the
data further before extracting the statistics that I am interested in.

It seems like my options are to either low pass filter the raw sensor
data further before computing the inverse solution, or to filter the
source data directly in matlab. Is there any methodological reason to
prefer one method over the other?

If filtering the source data in matlab is preferred (or if both methods
are fine), could anyone recommend a particular filter type? I've looked
into documentation for filtering in matlab and it seems like there are a
tremendous amount of options.

Thanks,

--
Matthew Panichello
Research Coordinator, Bar Group
Massachusetts General Hospital
Phone: 617-726-9034

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

-------------- next part --------------
A non-text attachment was scrubbed...
Name: eegfilt.m
Type: text/x-objcsrc
Size: 7563 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20121115/85521f3d/attachment.bin

3

Thanks Sheraz! The function works great.

Beyond computational cost, there's no reason to worry about filtering in
source space?

Thanks,

Matt

4

Hi Matt,
   For a given inverse operator, it is the same if you do the filtering in sensor or source space... The thing you might want to consider is if you want to make an inverse operator that is more tailored to the noise covariance that you get with the specific filtering you want to do. In that sense it is probably better at least on paper to filter in sensor space first and then generate a new noise covariance and inverse operator.

Regards,
Hari

Hari Bharadwaj
PhD Candidate, Biomedical Engineering,
Auditory Neuroscience Laboratory
Boston University, Boston, MA 02215

Martinos Center for Biomedical Imaging,
Massachusetts General Hospital
Charlestown, MA 02129

hari at nmr.mgh.harvard.edu
Ph: 734-883-5954

5

Hi Hari,

Thanks for clarifying. I'll try both methods on one of the ROIs and see
how much of a practical difference the filtering stage makes.

Thanks,

Matt

6

hi matt,

juste make sure you used pick_normal or fixed orientation so you don't
filter the signal
after taking an absolute value.

Best,
Alex

7

Hi Alex,

Thanks for letting me know. I specified --picknormcomp .

Thanks,

Matt