Dear Users,
Concatenation or higher level stats?
We would value the input of any users who have experience of 'combining' fMRI data across multiple runs for more robust averaging.
The issue arises when trials of an experient are acquired in different data blocks due to technical limitations of scan scquisition protocols or often to reduce the strain on participants in a single session.
I have searched many available resources. There are well documented routines for (1) analysinng the sessions individually and then compring them with a higher level analysis or (2) demeaning the two timeseries and combining them into a single one .. then following the standard analysis individual subject routine.
Comments would be appreciated (especially from anyone has first hand experience of doing this).
Thanks
Andre'
Hi Andre,
We are doing this at the moment with FSL. Our design has 10 subjects who each do 2 sessions (however, the number of sessions each subject does could vary). We do a first level analysis on the 20 (10 * 2) scans. We then do a second higher-level analysis in which we combine the two sessions from each subject. This is followed by a 3rd higher-level analysis in which we look at the activation across subjects for each contrast.
Cheers,
Tim
andre@ynic.york.ac.uk wrote:
Dear Users,
Concatenation or higher level stats?
We would value the input of any users who have experience of 'combining' fMRI data across multiple runs for more robust averaging.
The issue arises when trials of an experient are acquired in different data blocks due to technical limitations of scan scquisition protocols or often to reduce the strain on participants in a single session.
I have searched many available resources. There are well documented routines for (1) analysinng the sessions individually and then compring them with a higher level analysis or (2) demeaning the two timeseries and combining them into a single one .. then following the standard analysis individual subject routine.
Comments would be appreciated (especially from anyone has first hand experience of doing this).
Thanks
Andre'
Hi, Andre,
To continue with this issue. I think from a user point of view, it would be better to follow option (2) because you then do analyses as usual treating all the data for each subjects as you would if you did not have memory limitation.
If people have complicated designs, as it is normally the case for people who need to scan for a long time, the data analysis becomes even more complicated, and the room for errors increases.
I think I was wrong to suggest earlier that FSL would automatically deal with this problem. If I understand it correctly, the problem is that for different scans, the arbitrary intensity values are going to be very different, right?. So the mean intensity for one scan can be, say, 1000, while for another scan of the same individual, it could be 4000. Doing grand-mean scaling does not seem right here, which is what FSL automatically does. But if you de-mean each scan (i.e., subtract its raw mean intensity, which results in a mean intensity of 0 for each scan), there should be no problem then combining the scans as if they were all coming from the same one, as the mean intensity would be the same for all. The 'avwmaths' commands in FSL will compute this very easily (assuming I am getting this right).
Apparently, people did this routinely at the Waismann Institute, so I actually never heard of someone having to analyze the data separately for each scan. Maybe some softwares do this pre-processing by default. My experience with Voxbo was such that, even when we had 8 or 10 different scans per person (an hour of scanning), we did not have to worry about it in the analysis.
Silvia
On 8 Jul 2008, at 16:20, Tim Andrews wrote:
Hi Andre,
We are doing this at the moment with FSL. Our design has 10 subjects who each do 2 sessions (however, the number of sessions each subject does could vary). We do a first level analysis on the 20 (10 * 2) scans. We then do a second higher-level analysis in which we combine the two sessions from each subject. This is followed by a 3rd higher-level analysis in which we look at the activation across subjects for each contrast.
Cheers,
Tim
andre@ynic.york.ac.uk wrote:
Dear Users,
Concatenation or higher level stats?
We would value the input of any users who have experience of 'combining' fMRI data across multiple runs for more robust averaging.
The issue arises when trials of an experient are acquired in different data blocks due to technical limitations of scan scquisition protocols or often to reduce the strain on participants in a single session.
I have searched many available resources. There are well documented routines for (1) analysinng the sessions individually and then compring them with a higher level analysis or (2) demeaning the two timeseries and combining them into a single one .. then following the standard analysis individual subject routine.
Comments would be appreciated (especially from anyone has first hand experience of doing this).
Thanks
Andre'
-- Dr Tim Andrews Department of Psychology University of York York, YO10 5DD UK
Tel: 44-1904-434356 Fax: 44-1904-433181
http://www-users.york.ac.uk/~ta505/ http://www.york.ac.uk/depts/psych/www/admissions/cns/
-- ynic-users mailing list ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
Silvia Gennari Department of Psychology University of York York, YO10 5DD United Kingdom
Hi, again,
Just one more comment. What I just said only holds if you are scanning people in the same day one run after the other. If one scan people in different days, there are additional complications having to do with motion correction and alignment of the images over time. Are you talking about sessions in different days?
Silvia On 9 Jul 2008, at 12:11, Silvia Gennari wrote:
Hi, Andre,
To continue with this issue. I think from a user point of view, it would be better to follow option (2) because you then do analyses as usual treating all the data for each subjects as you would if you did not have memory limitation.
If people have complicated designs, as it is normally the case for people who need to scan for a long time, the data analysis becomes even more complicated, and the room for errors increases.
I think I was wrong to suggest earlier that FSL would automatically deal with this problem. If I understand it correctly, the problem is that for different scans, the arbitrary intensity values are going to be very different, right?. So the mean intensity for one scan can be, say, 1000, while for another scan of the same individual, it could be 4000. Doing grand-mean scaling does not seem right here, which is what FSL automatically does. But if you de-mean each scan (i.e., subtract its raw mean intensity, which results in a mean intensity of 0 for each scan), there should be no problem then combining the scans as if they were all coming from the same one, as the mean intensity would be the same for all. The 'avwmaths' commands in FSL will compute this very easily (assuming I am getting this right).
Apparently, people did this routinely at the Waismann Institute, so I actually never heard of someone having to analyze the data separately for each scan. Maybe some softwares do this pre- processing by default. My experience with Voxbo was such that, even when we had 8 or 10 different scans per person (an hour of scanning), we did not have to worry about it in the analysis.
Silvia
On 8 Jul 2008, at 16:20, Tim Andrews wrote:
Hi Andre,
We are doing this at the moment with FSL. Our design has 10 subjects who each do 2 sessions (however, the number of sessions each subject does could vary). We do a first level analysis on the 20 (10 * 2) scans. We then do a second higher-level analysis in which we combine the two sessions from each subject. This is followed by a 3rd higher-level analysis in which we look at the activation across subjects for each contrast.
Cheers,
Tim
andre@ynic.york.ac.uk wrote:
Dear Users,
Concatenation or higher level stats?
We would value the input of any users who have experience of 'combining' fMRI data across multiple runs for more robust averaging.
The issue arises when trials of an experient are acquired in different data blocks due to technical limitations of scan scquisition protocols or often to reduce the strain on participants in a single session.
I have searched many available resources. There are well documented routines for (1) analysinng the sessions individually and then compring them with a higher level analysis or (2) demeaning the two timeseries and combining them into a single one .. then following the standard analysis individual subject routine.
Comments would be appreciated (especially from anyone has first hand experience of doing this).
Thanks
Andre'
-- Dr Tim Andrews Department of Psychology University of York York, YO10 5DD UK
Tel: 44-1904-434356 Fax: 44-1904-433181
http://www-users.york.ac.uk/~ta505/ http://www.york.ac.uk/depts/psych/www/admissions/cns/
-- ynic-users mailing list ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
Silvia Gennari Department of Psychology University of York York, YO10 5DD United Kingdom
-- ynic-users mailing list ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
Silvia Gennari Department of Psychology University of York York, YO10 5DD United Kingdom
http://www.fmrib.ox.ac.uk/fslcourse/lectures/feat3.pdf, slide 30 shows how this can be done with FSL.
Tim
Silvia Gennari wrote:
Hi, again,
Just one more comment. What I just said only holds if you are scanning people in the same day one run after the other. If one scan people in different days, there are additional complications having to do with motion correction and alignment of the images over time. Are you talking about sessions in different days?
Silvia On 9 Jul 2008, at 12:11, Silvia Gennari wrote:
Hi, Andre,
To continue with this issue. I think from a user point of view, it would be better to follow option (2) because you then do analyses as usual treating all the data for each subjects as you would if you did not have memory limitation.
If people have complicated designs, as it is normally the case for people who need to scan for a long time, the data analysis becomes even more complicated, and the room for errors increases.
I think I was wrong to suggest earlier that FSL would automatically deal with this problem. If I understand it correctly, the problem is that for different scans, the arbitrary intensity values are going to be very different, right?. So the mean intensity for one scan can be, say, 1000, while for another scan of the same individual, it could be 4000. Doing grand-mean scaling does not seem right here, which is what FSL automatically does. But if you de-mean each scan (i.e., subtract its raw mean intensity, which results in a mean intensity of 0 for each scan), there should be no problem then combining the scans as if they were all coming from the same one, as the mean intensity would be the same for all. The 'avwmaths' commands in FSL will compute this very easily (assuming I am getting this right).
Apparently, people did this routinely at the Waismann Institute, so I actually never heard of someone having to analyze the data separately for each scan. Maybe some softwares do this pre-processing by default. My experience with Voxbo was such that, even when we had 8 or 10 different scans per person (an hour of scanning), we did not have to worry about it in the analysis.
Silvia
On 8 Jul 2008, at 16:20, Tim Andrews wrote:
Hi Andre,
We are doing this at the moment with FSL. Our design has 10 subjects who each do 2 sessions (however, the number of sessions each subject does could vary). We do a first level analysis on the 20 (10 * 2) scans. We then do a second higher-level analysis in which we combine the two sessions from each subject. This is followed by a 3rd higher-level analysis in which we look at the activation across subjects for each contrast.
Cheers,
Tim
andre@ynic.york.ac.uk mailto:andre@ynic.york.ac.uk wrote:
Dear Users,
Concatenation or higher level stats?
We would value the input of any users who have experience of 'combining' fMRI data across multiple runs for more robust averaging.
The issue arises when trials of an experient are acquired in different data blocks due to technical limitations of scan scquisition protocols or often to reduce the strain on participants in a single session.
I have searched many available resources. There are well documented routines for (1) analysinng the sessions individually and then compring them with a higher level analysis or (2) demeaning the two timeseries and combining them into a single one .. then following the standard analysis individual subject routine.
Comments would be appreciated (especially from anyone has first hand experience of doing this).
Thanks
Andre'
-- Dr Tim Andrews Department of Psychology University of York York, YO10 5DD UK
Tel: 44-1904-434356 Fax: 44-1904-433181
http://www-users.york.ac.uk/~ta505/ http://www-users.york.ac.uk/%7Eta505/ http://www.york.ac.uk/depts/psych/www/admissions/cns/
-- ynic-users mailing list ynic-users@ynic.york.ac.uk mailto:ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
Silvia Gennari Department of Psychology University of York York, YO10 5DD United Kingdom
-- ynic-users mailing list ynic-users@ynic.york.ac.uk mailto:ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
Silvia Gennari Department of Psychology University of York York, YO10 5DD United Kingdom
Thank you all for your invaluable input.
We will now revise the options with the users so that they can decide which path they would like to take.
This is a fantastic example of how we might call upon our local expertise to inform users and help to minimise duplication of work.
We will close this mailing item with a statement about the options that were chosen for analysis and then transfer the lot to an FAQ list.
Thank you again for your input and enthusiasm.
Andre'
http://www.fmrib.ox.ac.uk/fslcourse/lectures/feat3.pdf, slide 30 shows how this can be done with FSL.
Tim
Silvia Gennari wrote:
Hi, again,
Just one more comment. What I just said only holds if you are scanning people in the same day one run after the other. If one scan people in different days, there are additional complications having to do with motion correction and alignment of the images over time. Are you talking about sessions in different days?
Silvia On 9 Jul 2008, at 12:11, Silvia Gennari wrote:
Hi, Andre,
To continue with this issue. I think from a user point of view, it would be better to follow option (2) because you then do analyses as usual treating all the data for each subjects as you would if you did not have memory limitation.
If people have complicated designs, as it is normally the case for people who need to scan for a long time, the data analysis becomes even more complicated, and the room for errors increases.
I think I was wrong to suggest earlier that FSL would automatically deal with this problem. If I understand it correctly, the problem is that for different scans, the arbitrary intensity values are going to be very different, right?. So the mean intensity for one scan can be, say, 1000, while for another scan of the same individual, it could be 4000. Doing grand-mean scaling does not seem right here, which is what FSL automatically does. But if you de-mean each scan (i.e., subtract its raw mean intensity, which results in a mean intensity of 0 for each scan), there should be no problem then combining the scans as if they were all coming from the same one, as the mean intensity would be the same for all. The 'avwmaths' commands in FSL will compute this very easily (assuming I am getting this right).
Apparently, people did this routinely at the Waismann Institute, so I actually never heard of someone having to analyze the data separately for each scan. Maybe some softwares do this pre-processing by default. My experience with Voxbo was such that, even when we had 8 or 10 different scans per person (an hour of scanning), we did not have to worry about it in the analysis.
Silvia
On 8 Jul 2008, at 16:20, Tim Andrews wrote:
Hi Andre,
We are doing this at the moment with FSL. Our design has 10 subjects who each do 2 sessions (however, the number of sessions each subject does could vary). We do a first level analysis on the 20 (10 * 2) scans. We then do a second higher-level analysis in which we combine the two sessions from each subject. This is followed by a 3rd higher-level analysis in which we look at the activation across subjects for each contrast.
Cheers,
Tim
andre@ynic.york.ac.uk mailto:andre@ynic.york.ac.uk wrote:
Dear Users,
Concatenation or higher level stats?
We would value the input of any users who have experience of 'combining' fMRI data across multiple runs for more robust averaging.
The issue arises when trials of an experient are acquired in different data blocks due to technical limitations of scan scquisition protocols or often to reduce the strain on participants in a single session.
I have searched many available resources. There are well documented routines for (1) analysinng the sessions individually and then compring them with a higher level analysis or (2) demeaning the two timeseries and combining them into a single one .. then following the standard analysis individual subject routine.
Comments would be appreciated (especially from anyone has first hand experience of doing this).
Thanks
Andre'
-- Dr Tim Andrews Department of Psychology University of York York, YO10 5DD UK
Tel: 44-1904-434356 Fax: 44-1904-433181
http://www-users.york.ac.uk/~ta505/ http://www-users.york.ac.uk/%7Eta505/ http://www.york.ac.uk/depts/psych/www/admissions/cns/
-- ynic-users mailing list ynic-users@ynic.york.ac.uk mailto:ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
Silvia Gennari Department of Psychology University of York York, YO10 5DD United Kingdom
-- ynic-users mailing list ynic-users@ynic.york.ac.uk mailto:ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
Silvia Gennari Department of Psychology University of York York, YO10 5DD United Kingdom
-- Dr Tim Andrews Department of Psychology University of York York, YO10 5DD UK
Tel: 44-1904-434356 Fax: 44-1904-433181
http://www-users.york.ac.uk/~ta505/ http://www.york.ac.uk/depts/psych/www/admissions/cns/
-- ynic-users mailing list ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
Hi Andre,
I have done that many times at the Montreal Neurological Institute with fMRIStat (the MNI software). For example, in 1 experiment I had 6 runs per subject. So, I analyzed each run separately first specifying all the contrasts and then combined all 6 runs of each subject together at a second level of analysis. Mind you, all the conditions/contrasts appeared in all the runs. Then transformed those results into Talairach space before doing the group analysis of all the subjects (I had 15 participants).
I hope this helps, Katerina
**************************************************************** Dr. Ekaterini Klepousniotou Lecturer in Cognitive Neuroscience & Neuropsychology Institute of Psychological Sciences University of Leeds Leeds LS2 9JT UK Tel: +44 (0)113 3435716 Fax: +44 (0)113 3435749
-----Original Message----- From: andre@ynic.york.ac.uk [mailto:andre@ynic.york.ac.uk] Sent: 08 July 2008 13:46 To: ynic-users@ynic.york.ac.uk Subject: Call for expertise: across session fMRI analyses
Dear Users,
Concatenation or higher level stats?
We would value the input of any users who have experience of 'combining' fMRI data across multiple runs for more robust averaging.
The issue arises when trials of an experient are acquired in different data blocks due to technical limitations of scan scquisition protocols or often to reduce the strain on participants in a single session.
I have searched many available resources. There are well documented routines for (1) analysinng the sessions individually and then compring them with a higher level analysis or (2) demeaning the two timeseries and combining them into a single one .. then following the standard analysis individual subject routine.
Comments would be appreciated (especially from anyone has first hand experience of doing this).
Thanks
Andre'