Dear Andrew,
On Wed, May 22, 2019 at 11:34 AM Andrew Andy <[log in to unmask]> wrote:
>
> Dear Vladimir,
>
> Thank you so much for your kindness. I tried to implement your advice, but I am a little confused.
> 1- After applying this solution, there is no need to do source reconstruction? That means the newly constructed D.data is the signal of the sources?
The virtual montage mechanism works with the @meeg object. D.data is
not changed but when you load the object with D = spm_eeg_load and
then index D as an array the montage is applied on the fly and it
looks as if you have source data.
> 2- For constructing montage matrix, I need to make a matrix with dimension of "number of sources * number of electrodes", for my data set this is 20484*63. The projector U is constant for all of the sources and the code uses the same matrix (dimension of 59*63) for all vertexes of the mesh. So how should I concatenate the projectors to make a meaningful montage matrix? Should I include inverse.M too?
Yes, sorry my mistake. The projector should be M*U.
Best,
Vladimir
>
> Best Regards,
> Andrew
>
> On Mon, May 20, 2019 at 1:28 PM Vladimir Litvak <[log in to unmask]> wrote:
>>
>> Dear Andrew,
>>
>> As long as your source data are a linear transformation of the sensor
>> data you could use the virtual montage functionality in SPM. You would
>> have to modify the code of spm_eeg_inv_extract to concatenate the
>> projectors U (around line 149) and aggregate them in a montage struct
>> (see section
>> 12.8.6 of the manual) and then you could add this montage to your
>> dataset (try 'help meeg/montage). I'm not sure it makes a lot of
>> sense to do that in combination with a sparse solution such as MSP
>> because most of your source waveforms will be zero. But if you use IID
>> or COH options it should be OK.
>>
>> Best,
>>
>> Vladimir
>>
>> On Mon, May 20, 2019 at 8:22 AM Andrew Andy <[log in to unmask]> wrote:
>> >
>> > Hi Vladimir,
>> >
>> > Thank you for your clear response. I want to extract all sources' signal with resolution of 3mm. Firstly, I tried radius=3mm, but it took a long time to get all of the VOI's signal. However, extracting multiple sources with radius of 0mm takes time as long as one source's signal extraction. I was thinking to use the latter one (extracting signal of each vertex and then averaging). But apparently, such process needs large amounts of RAM and also high-capacity hard drive to save the results.
>> > Do you have any suggestion to do source extraction for whole brain more efficiently?
>> > Thank you so much for your consideration.
>> >
>> > Best,
>> > Andrew
>> >
>> > On Wed, May 15, 2019 at 12:15 AM Vladimir Litvak <[log in to unmask]> wrote:
>> >>
>> >> Dear Andrew,
>> >>
>> >> We would normally use 3mm radius. If there are multiple vertices in the ROI the extraction code computed the first principal component rather than the mean.
>> >>
>> >> Best,
>> >>
>> >> Vladimir
>> >>
>> >> > On 14 May 2019, at 12:24, Andrew Andy <[log in to unmask]> wrote:
>> >> >
>> >> > Hi dear all,
>> >> > I am using "source extraction" for EEG data. I am confused with the effect of radius for VOI on source's signal.
>> >> > For extracting the signal of a VOI with radius of 3mm, which procedure is more accurate?
>> >> >
>> >> > 1- Defining one source with radius of 3mm.
>> >> > or
>> >> > 2- Defining the vertexes that are in the VOI as separate sources with radius of zero and averaging over sources to get one signal for the whole VOI.
>> >> >
>> >> > And why two signals are different?
>> >> >
>> >> > Any help would be appreciated.
>> >> >
>> >> > Best,
>> >> > Andrew
>> >> >
|