Dear SPMers:
This follow my previous (too long) email where I did not get any reply.
I did a correlation study (PET on n=42, binding vs, con [1 0] and
[-1 0] for pos and neg correlations) with SPM96 and also SPM99 using
exactly same subjects, design
I obtained exactly the same regions and Z values.
I extracted voxel values in SPM96 (results, plot and then "y") and in
SPM99 (VOI, no adjustement for contrast for the data and xY.y or
equivalently Y since corrcoef(Y, xY.y) = 1)
The % change shown by plot of y (SPM96) against age
was significantly different (2 -3 time larger with SPM99 depending on
regions) from the
% change shown by plot of xY.y (or Y) in SPM99 against age.
In SPM99, y are the ADJUSTED data (blue) and Y the FITTED data (red)
AND in SPM96, y are the ADJUSTED data. I found, however that
Y(SPM99) the FITTED data (or xY.y) and NOT y(SPM99) the ADJUSTED
data displayed the highest correlation with y(SPM96) the ADJUSTED
data which ranged between 0.61 and 0.79.
Can someone explain why I cannot replicate the % change I found with SPM96
and equivalently how can I get the % change in SPM99. After looking
at SPM help I was not able to find the answer.
Thanks for your help
PS: If interested by details, below is my previous email after
correcting a mistake in region 2
Dear SPMers;
I analyzed the same set of data (n=42, correlation binding with age)
using SPM96 and SPM99 (my purpose was NOT a cross validation but I
wanted to report these results and others using a similar statistical
package SPM99). By doing that I realized that the % change I was
computing with SPM96 was almost half (1.74 less) than the one I
computed with SPM99.
I found that the intercept (b)and slope (m) (computed with Matlab)
were different between SPM96 (residual d.f. = 37) and SPM99 (residual
d.f. = 37) but were highly correlated.
The slopes and intercept obtained on 8 regions (placed on Z max, 7
regions of positive correlations and 1 negative correlation) with
SPM96 and SPM99 are:
m99 = 1.7473 * m96 + 39.0830, r = 0.9965 AND b99 =3.1188* b96 -
2.9969, r= 0.9616.
Question A: is what slope and Intercept should I use to compute %
change in binding versus change in age since there is a 1.7 fold
change in slope between both SPM96 and SPM99. I guess that I should
use SPM99 but then why this very CONSISTENT discrepancy between SPM96
and SPM99.
Question B: why there is a high correlation between the m's
obtained with SPM96 and 99 (and between the b's).
Question C: The correlation coefficient for relatively low Z ( region
6 and 7, 4.23 and 3.92 ) or high Z( region 8, 7.38 and 7.77) shows
larger discrepancies compared to the very good reproducibility for
other regions although for region 8 there is also a large difference
in the computed Z value between SPM96 and SPM99 (but not enough (?)
to explain the differences in r) .
Below are the actual results.
POSITIVE CORRELATION
SPM96 and SPM99 showed exactly the same 7 regions, exactly the same Z
, same coordinates for max, minor discrepancy in the size of the
clusters in 3 of 7 regions.
I extracted voxel values in SPM96 ( results, plot and then "y") and
in SPM99 (VOI, no adjustement for contrast for the data and xY.y
since I found that xY.y obtained using no adjustement had the highest
correlation with SPM96 "y").
data are shown below as: SPM version (96 or 99) / Size k/ Z max/
Pearson correlation/ Slope (m) /Intercept 9b)
region1
96 k=752, Z=4.84, r= 0.7621, m=0.0106, b=1.0893
99 k=747, Z=4.84, r= 0.7523, m=0.0203, b=0.6796
region2
96 k=171, Z=4.45, r= 0.7008, m=0.0060, b=1.1782
99 k=171, Z=4.45, r= 0.6994, m=0.0145, b=0.8206
region3
96 k=239, Z=4.42, r= 0.7036, m=0.0071, b=1.3256
99 k=239, Z=4.42, r= 0.7068, m=0.0176, b=0.8807
region4
96 k=341, Z=4.41, r= 0.7144, m=0.0078, b=1.2544
99 k=330, Z=4.41, r= 0.7098, m=0.0174, b=0.8479
region5
96 k=671, Z=4.27, r= 0.6947, m=0.0067, b=1.3630
99 k=671, Z=4.27, r= 0.6903, m=0.0179, b=0.8887
region6
96 k=202, Z=4.23, r= 0.6510, m=0.0063, b=1.1039
99 k=201, Z=4.23, r= 0.6978, m=0.0152, b=0.73.07
region7
96 k=787, Z=3.92, r= 0.6244, m=0.0084, b=1.3569
99 k=748, Z=3.92, r= 0.6867, m=0.0173, b=0.9820
NEGATIVE CORRELATION
SPM96 and SPM99 showed the same large region
region8
96 k=8133, Z=7.38, r=-0.9199, m=-0.0186, b=2.0336
99 k=8101, Z=7.77, r=-0.7667, m=-0.0290, b=3.5813
--
Badreddine Bencherif, MD
Department of Radiology
Division of Nuclear Medicine
Johns Hopkins University School of Medicine
601 N. Caroline St. / JHOC 4230
Baltimore, MD 21287-0855
Phone : (410) 614-2787
Pager : (410) 283-2050
Fax : (410) 614-1977
email : [log in to unmask]
|