Brandall Y. Suyenobu wrote:

> Hello SPM users
>
> Is there a consensus as to the optimal way of analyzing multiple fMRI
> runs of a single subject as separate sessions?


Probably not ! But I'll offer my opinion  !

> In a recent post the
> separate session analysis was pointed out as the "normal procedure"
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0302&L=spm&P=R33056&I=-3
> <http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0302&L=spm&P=R33056&I=-3>,
> and previous posts (e.g.,
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0207&L=spm&P=R13130&I=-3)
> have pointed out that, although valid, across-session contrasts are
> usually less statistically efficient;


I think Rik's (Henson) point here is
that across-session contrasts are less
statistically efficient because the
effects are further away in time - and
therefore will not be so well matched
(matched comparisons being more
sensitive than unmatched ones).

But this point is a critique of the
design - not on how the data are
analysed ! ie. if you model the sessions
separately or together it won't make the
effects any closer together in time.

> however, in the case of the
> multiple-run-single-subject analysis, there seems to be little
> alternative with some experimental designs.  In a current study we are
> obtaining 6 10-min scan series (runs) per subject per session, and
> within each run presenting the subject with 3 active conditions and 2
> rest conditions in a total of 16 presentations. When analyzing this as
> separate sessions, is it best to employ second level analyses with the
> relevant contrasts when, e.g., summarizing condition effects over all 6
> runs?
>


If you have a multiple subject -
multiple session (per subject) study
then it is possible to do RFX analysis
both over sessions and then over subjects.

If you did do this then you would be
applauded for your rigor but I think
most people would do FFX over sessions
and then RFX over subjects - this is
because the between-subject variance is
likely to be larger than the
between-session variance (although this
is an empirical issue related to a
number of factors including eg.
within-session sample size).


If, however, you are doing a single
subject (or a small number of subjects
eg. 3) case-study then RFX over sessions
is a reasonable option.


> And, alternatively, is there an acceptable (i.e., least problematic)
> method of analyzing multiple runs of a single subject as a single
> session?  In our study, the TR of 3 sec yields 200 scans per run that I
> have analyzed as a single session (fixed response/box-car analysis),
> 1200 scans, with, I think, reasonable results.

> As I am not modeling the
> rest conditions, the sum of the time between analyzed segments exceeds
> the inter-session interval  does this have a bearing on the
> appropriateness of a single-session analysis?  Any comments will be most
> welcomed.
>
> Best regards,
>
> Brandall Y. Suyenobu, Ph.D.
> Staff Research Associate
>
> Brain Imaging Group
> CNS: Center for Neurovisceral Sciences & Women's Health
> CURE: Digestive Diseases Research Center
> UCLA Division of Digestive Diseases
> Greater Los Angeles Veterans' Administration Health Care,
> Bldg. 115, rm 223
> 11301 Wilshire Boulevard
> Los Angeles, California 90073
> Tel: (310) 478-3711 ext. 40580
> Fax: (310) 794-2864
> http://ibs.med.ucla.edu <http://ibs.med.ucla.edu/>
> http://mindbody.med.ucla.edu <http://mindbody.med.ucla.edu/>
>


--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG

Tel: 020 7833 7478
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/