HI Xiaozhen,

> According to your opinion, I did not use gender as a covariate any more.

If you prefer, you can still model the gender effect with 3dttest++ with both options -setA and -setB.

> 3dttest++ -setA subjects_image_file -covariates covariate_file -prefix test_result

Did you center your covariates? Read the help of 3dttest++ for the details regarding centering:

3dttest++ -help | less

> The results include 6 sub-bricks, setA mean, setA Tstat, setA age-mean, setA age-Tstat,
> setA blood-test-mean, setA blood-test-Tstat. How to explain the result based on what I want?

The first two sub-bricks ( setA mean, setA Tstat) are for the group effect controlled at the center values of the two covariates. The next two (setA age-mean, setA age-Tstat) are for the marginal effect (slope) of age. And the last two are about the slope of blood test.

> I assume that every subject is a time point of whole group when I use 3dDeconvolve to remove
> age effect, then I use 3dTcorr1D to get the correlation coefficient between residual and blood-test
> result. Is that okay?

I'm not so sure why you're talking about time point. You referred in your original post the z-score from some other analysis at individual level. So I've been assuming that you're taking one z-score volume for group analysis; and 3dttest++ would give you all you want. Why bother 3dDeconvolve + 3dTcorr1D? Unless you're talking about something else...

Gang