Dear Vladimir,
I tried to make some 19-channel synthetic EEG data by considering just two voxels to be active. I just assumed that the signal of each voxel is a Gaussian window. Then, using the same forward matrix as my real data, I computed the output as LJ (without applying any temporal or spatial correlations). Then, added some Gaussian random noise to each channel. The norm of the noise in each channel is one-tenth of the norm of signal in that channel. I did source reconstruction on my data in 3 different situations:
1. Source reconstruction on data before adding noise and considering just one trial
2. Source reconstruction on data after adding noise and considering just one trial
3. Source reconstruction on data after adding noise and considering 100 trials. (The difference between the trials is made by different noise signals)
Obviously, I converted my data to SPM8 by changing the scripts.
The attached file is the results of these three reconstructions. In the file, you can find the exact location of those two voxels. As you can see, it can estimate the position of one of the voxels but it gets confused for the other voxel. Also, for the frontal voxel, it can recognize that it is a unilateral source but for the other voxel it can never recognize that.
I am very confused right now. Do you think I am doing something wrong or I have to consider some other assumptions? I have seen your paper on MSP and how you produce synthetic data but I just want to know why it does not work on my data? Can it be because of those temporal and spatial correlations you apply to your synthetic data?
Best,
Pegah
--------------------------------------------------------
Pegah Tayaranian Hosseini
PhD Student
Room 4077, Tizard building (13)
Institute of Sound and Vibration Research
University of Southampton, SO17 1BJ, UK
Tel: 023 8059 2850
email: [log in to unmask]
|