Unfortunately, I can't help you solve this "end-droop" problem, but I have
encountered it in previous studies. In unpublished work, we looked at trial
by trial classification of faces, objects, scenes, etc. in the visual cortex
(similar to Haxby, and Cox and Savoy), and found that classification
accuracy was worse at each end of the scan session. Closer inspection of
the data showed exactly the effect that you have described. In our case, we
just left out the end data in order to get a better measure of
classification accuracy.
- Paul
Quoting "Andrew J. Gerber" <[log in to unmask]>:
> Dear SPMers,
>
> I am hoping that someone can help me understand how to interpret an
> intermediate variable in the function spm_spm when applying the AR(1)
> model for an estimation of fMRI data. To my reading of the code, on
> line 614 (line numbers are specified for SPM2), raw BOLD data is
> loaded into the matrix Y:
> Y(i,Cm = spm_get_data(VY(i),xyz(:,Cm));
>
> Then on line 638 of the second pass of this function (when using an
> AR(1) model), the raw data undergoes high pass filtering and AR(1)
> correction:
> KWY = spm_filter(xX.K,W*Y);
>
> And finally on line 644, the least squares regression is calculated:
> beta = xX.pKX*KWY;
>
> If this is correct, shouldn't the variable KWY contain the filtered
> BOLD data that is used to compute the least squares regression? When
> I plot KWY against scan number (273 scans) for any of my subjects I
> see the following seemingly strange features: (1) the first and last
> points are much larger than the rest of the data (see first figure),
> (2) if I remove the first and last points manually, I see a subtle
> pattern of an upward linear drift in the first 40 or so scans and a
> downward linear drift in the last 40 scans (see second figure). This
> occurs for all my subjects and runs.
>
>
> []
>
>
> []
>
>
> Has anyone else seen this pattern? Am I misinterpreting the variable
> KWY? (if so, what intermediate variable contains the filtered BOLD
> data?) Does this suggest something wrong with my underlying data or
> the way the AR(1) model is being applied here? I see the above effect
> only when AR(1) is in the model and in both SPM2 and SPM5 using the
> classical estimation procedure.
>
> Thanks in advance for your help.
>
> Best, Andrew
>
>
> Andrew J. Gerber, MD, PhD
> Post-doctoral Research Fellow
> Columbia University/NYSPI
> Division of Child and Adolescent Psychiatry
> 1051 Riverside Drive, Unit 74
> New York, New York 10032
> Email: [log in to unmask]
> Phone: (212) 543-5104; Fax: (212) 543-0522
--
Paul K. Mazaika, PhD
Center for Interdisciplinary Brain Sciences Research
Stanford University School of Medicine
401 Quarry Road, MC 5795
Stanford, Ca 94305
Office: (650)724-6646 Cell: (650)799-8319
CONFIDENTIALITY NOTICE: Information contained in this message and any
attachments is intended only for the addressee(s). If you believe
that you have received this message in error, please notify the
sender immediately by return electronic mail, and please delete it
without further review, disclosure, or copying
|