2009/7/2 hiten <[log in to unmask]>:
> Dear SPMers:
> Now I am confused about the filtering of M/EEG data.The data was collected
> by VSM CTF275 MEG system,sample at 1200HZ,under incongruent color-words
> STROOP task.The data are preprocessed:
> convert,epoching(-200ms--800ms),aritifacts(1*e-12T).the attached pictures
> has the follwing meanings:
> Pic 1 mean: after aritifact then average signals of all
> 64 trials
> Pic 1mean+2filter: filtering the averaged signals using the
> butterworth bandpass(1-100HZ)
> Pic 1filter+2mean firstly filter(butterworth bandpass(1-100HZ)),then
> Pic 1mean+2wt use my improved wavelet-based de-noising
> algorithm to the averaged signals
> 'Mechanisms of evoked and induced responses in MEG/EEG',Olivier David, James
> M. Kilner,* and Karl J. Friston.NeuroImage 31 (2006) 1580 - 1591 clearly
> shows that evoked responses are determinate,time/wave locked. so I have the
> following questions:
> 1.how to set the bandpass width accurately? could somebody give me some
That depends on your data and on the purpose of your analysis. In
principle you should select the bandpass in such a way that the
activity of interest will be included and noise will be excluded.
> 2.which is the best processing sequences: first average then filter or first
> filter then average? However, no matter Pic1mean+2filter or Pic1filter+2mean
> are deformated comparing to Pic 1 mean ,which I think is not in conformity
> with the actual situation.In my opinion,filtering should be done using a
> zero-phase band-pass filter while spm8 not.
In principle both are possible. I'm surprised that there is such a big
difference between the results and it looks like the original data is
distorted too much. SPM does use a zero-phase bandpass filter but
perhaps for such a wide band, you should use 1Hz highpass and 100Hz
lowpass applied sequentially rather than a bandpass. Another
possibility that many people use especially for low-cutoff highpass
filter is to filter continuous data before epoching to avoid edge
> 3.In theory, Pic 1 mean is the evoked respones with additive gaussian noise.
> Pic 1mean+2wt is most similar to Pic 1 mean. So I think Pic 1mean+2wt is
> most closet to the realistic situation. Am I right?
It looks quite good but you are not writing just to advertise your
denoising method, are you ;-) ?
> Any comments would be greatly appreciated. Thanks in advance!
> haiteng jiang
> Research Center for Learning Science,
> Southeast University
> Si Pai Lou 2 # , Nanjing, 210096, P.R.China
> Brain Imaging Lab
> Email: [log in to unmask]