Re: [HCP-Users] MEG resting state

[Georgios Michalareas]

Hi Adoney,

you can just get the channel positions of the same sensors in order to 
compare them.


I am pasting here some code that loads the channel positions from the 
three different resting state scans in a common order and plots the 
histogram of the position differences between runs 1-2,1-3 and 2-3.


At first it does this for all the sensors including the reference sensors.

Then it does the same only for the MEG sensors , with the reference 
sensors removed, because the reference sensors are far from the  head 
and a small head movement can be translated into larger position 
differences further away from the head as compared to the MEG sensors 
which are near the head.


In the code below you need to change the data directory and Fieltrip 
directory to what you have.


I hope it helps

Best

Giorgos

%-
%fieldtripdir='/home/michalareasg/toolboxes/matlab/fieldtrip/fieldtrip_svn';
fieldtripdir='/home/georgios.michalareas/workspace/toolboxes/matlab/fieldtrip';
addpath(fieldtripdir);
ft_defaults;
%-

datadir='/home/georgios.michalareas/workspace/projects/HCP/data/connectomedb/177746/MEG/Restin/rmegpreproc/';

rsfile1=[datadir,'177746_MEG_3-Restin_rmegpreproc.mat'];
rsfile2=[datadir,'177746_MEG_4-Restin_rmegpreproc.mat'];
rsfile3=[datadir,'177746_MEG_5-Restin_rmegpreproc.mat'];

%=
load(rsfile1,'data');
data1=data; clear data;
%
load(rsfile2,'data');
data2=data; clear data;
%
load(rsfile3,'data');
data3=data; clear data;
%=

%%


[C12,IA12,IB12]=intersect(data1.grad.label,data2.grad.label);
[C13,IA13,IB13]=intersect(data1.grad.label,data3.grad.label);
if ~isequal(C12,C13)
    error('common grad labels between resting state recordins 1-2 and 
recordings 1-3 appear to be different')

else
    C=C12;
    IA1=IA12;
    IA2=IB12;
    IA3=IB13;
end

gradlabels=C;
gradpos1=data1.grad.chanpos(IA1,:);
gradpos2=data2.grad.chanpos(IA2,:);
gradpos3=data3.grad.chanpos(IA3,:);

posdif12=sqrt(sum((gradpos1-gradpos2).^2,2));
posdif13=sqrt(sum((gradpos1-gradpos3).^2,2));
posdif23=sqrt(sum((gradpos2-gradpos3).^2,2));

figure;
subplot(1,3,1);
hist(posdif12);
subplot(1,3,2);
hist(posdif13);
subplot(1,3,3);
hist(posdif23);



%% 
% Now do the same for only the MEG sensors without the REFERENCE sensors
% because they are too far from the head an a small head movement can be
% translated into bigger position differences of the reference sensors as
% compared to the MEG sensors which are close to the head

[chansMEG] = ft_channelselection('MEG',data1.grad.label );


[C1,IA1,IB1]=intersect(chansMEG, data1.grad.label);
[C2,IA2,IB2]=intersect(chansMEG, data2.grad.label);
[C3,IA3,IB3]=intersect(chansMEG, data3.grad.label);


if (~isequal(C1,C2))|(~isequal(C1,C3))|(~isequal(C2,C3))
    error('common grad labels between resting state recordins 1-2 and 
recordings 1-3 appear to be different')

else
    C=C1;
end

gradlabels=C;
gradpos1=data1.grad.chanpos(IB1,:);
gradpos2=data2.grad.chanpos(IB2,:);
gradpos3=data3.grad.chanpos(IB3,:);

posdif12=sqrt(sum((gradpos1-gradpos2).^2,2));
posdif13=sqrt(sum((gradpos1-gradpos3).^2,2));
posdif23=sqrt(sum((gradpos2-gradpos3).^2,2));

figure;
subplot(1,3,1);
hist(posdif12);
subplot(1,3,2);
hist(posdif13);
subplot(1,3,3);
hist(posdif23);









On 11/19/2017 9:40 PM, A Nunes wrote:

Hi Giorgos,

Thanks, I did not know about ft_megrealign. Idially I would like to 
just concatenate the data in the sensor level. I recently realized 
that between recordings the sensor position varies up to 0.2 meters in 
data.grad.chanpos. This is because the order of the sensors are 
different between recordings, I don't know why though. In 
data.grad.labelorg it shows the right order, however I have the 
impression that the order of the sensors in the covariance matrix is 
based in chanpos because separate recording have smaller rank than 
concatenating which has full rank. When reading the data file FT gives 
me the warning 'Your data and configuration allow for multiple sensor 
definitions.' maybe this is related.


Do you know how to approach this? Maybe I should somehow specify that 
the right order is in data.grad.labelorg rather than data.grad.label?


Thanks
Adonay

On Fri, Nov 17, 2017 at 5:15 AM, Michalareas, Georgios > wrote:


Hi Adoney,
The released data has been only 7.5 mm maximum movement within scan.
So the sensor positions should be quite close in all three resting
state scans.
ICA was done with all three sessions concatenated together.
So I believe it should be one to concatenate them for your
analysis too and assume sensors are more or less on same location.
If you want to treat each different scan separately then you have
to follow different strategies like:
-Perform beamforming for each session separately and project 

Re: [HCP-Users] MEG resting state

[A Nunes]

Hi Giorgos,

Thanks, I did not know about ft_megrealign. Idially I would like to just
concatenate the data in the sensor level. I recently realized that between
recordings the sensor position varies up to 0.2 meters in
data.grad.chanpos. This is because the order of the sensors are different
between recordings, I don't know why though. In data.grad.labelorg it shows
the right order, however I have the impression that the order of the
sensors in the covariance matrix is based in chanpos because separate
recording have smaller rank than concatenating which has full rank. When
reading the data file FT gives me the warning 'Your data and configuration
allow for multiple sensor definitions.' maybe this is related.

Do you know how to approach this? Maybe I should somehow specify that the
right order is in data.grad.labelorg rather than data.grad.label?

Thanks
Adonay

On Fri, Nov 17, 2017 at 5:15 AM, Michalareas, Georgios 
wrote:

> Hi Adoney,
> The released data has been only 7.5 mm maximum movement within scan.
> So the sensor positions should be quite close in all three resting state
> scans.
> ICA was done with all three sessions concatenated together.
> So I believe it should be one to concatenate them for your analysis too
> and assume sensors are more or less on same location.
> If you want to treat each different scan separately then you have to
> follow different strategies like:
> -Perform beamforming for each session separately and project each session
> separately into source space and then concatenate data in source space and
> do further analysis there.
> -Make a virtual gradiometer array as the mean of the three gradiometer
> positions in the three scans and then interpolate the data of each session
> to this virtual MEG sensors using Fieltrip’s function ft_megrealign and
> then combine all data in this virtual sensor space. Keep in mind that this
> projects the data into source space and projects it back to the virtual
> sensor space.
>
> I hope this helps
> Best
> Giorgos
>
> From:  on behalf of A Nunes <
> adonay.s.nu...@gmail.com>
> Date: Thursday, 16. November 2017 at 21:56
> To: "hcp-users@humanconnectome.org" 
> Subject: [HCP-Users] MEG resting state
>
> Hi,
>
> The MEG resting state data is split in three sessions, is it possible to
> append the data before computing the covariance matrix?
> I have some doubts because the sensor position might change between the
> recordings and if ICA was done separately, then the rank would change
> between sessions and I don't know how would this affect beamforming.
>
> Any suggestions?
>
> Thanks
> Adonay
>
> ___
> HCP-Users mailing list
> HCP-Users@humanconnectome.org
> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>

___
HCP-Users mailing list
HCP-Users@humanconnectome.org
http://lists.humanconnectome.org/mailman/listinfo/hcp-users

Re: [HCP-Users] MEG resting state

[Michalareas, Georgios]

Hi Adoney,
The released data has been only 7.5 mm maximum movement within scan.
So the sensor positions should be quite close in all three resting state scans.
ICA was done with all three sessions concatenated together.
So I believe it should be one to concatenate them for your analysis too and 
assume sensors are more or less on same location.
If you want to treat each different scan separately then you have to follow 
different strategies like:
-Perform beamforming for each session separately and project each session 
separately into source space and then concatenate data in source space and do 
further analysis there.
-Make a virtual gradiometer array as the mean of the three gradiometer 
positions in the three scans and then interpolate the data of each session to 
this virtual MEG sensors using Fieltrip’s function ft_megrealign and then 
combine all data in this virtual sensor space. Keep in mind that this projects 
the data into source space and projects it back to the virtual sensor space.

I hope this helps
Best
Giorgos

From: 
>
 on behalf of A Nunes 
>
Date: Thursday, 16. November 2017 at 21:56
To: "hcp-users@humanconnectome.org" 
>
Subject: [HCP-Users] MEG resting state

Hi,

The MEG resting state data is split in three sessions, is it possible to append 
the data before computing the covariance matrix?
I have some doubts because the sensor position might change between the 
recordings and if ICA was done separately, then the rank would change between 
sessions and I don't know how would this affect beamforming.

Any suggestions?

Thanks
Adonay

___
HCP-Users mailing list
HCP-Users@humanconnectome.org
http://lists.humanconnectome.org/mailman/listinfo/hcp-users

___
HCP-Users mailing list
HCP-Users@humanconnectome.org
http://lists.humanconnectome.org/mailman/listinfo/hcp-users