> I would like to know what is the difference of realignment process between
> SPM2 and SPM5.
>
> As I know, there is no algorithm difference implemented in SPM2 and SPM5.
>
> For instance, when I saw "spm_realign.m", the implementations are same.
The implementations are virtually identical. The only differences in the code
are:
1) Documentation changes, and more informative error messages.
2) Additional flexibility in the algorithm (to allow the possibility of 12
parameter registration - which may have been a request on the list).
3) Use of &&, rather than & (because it is supposed to be marginally faster
(actually because the "mlint" comand suggested it).
>
> However, when I tried to realign the same dataset with SPM2 and SPM5, SPM2
> is faster than SPM5.
>
> What makes this duration taken difference?
Timing differences are a result to a change in spm_defaults.m. In spm2, the
file said:
defaults.realign.estimate.quality = 0.75;
whereas in spm5, it says:
defaults.realign.estimate.quality = 0.9;
This setting determines which voxels are used for estimating the movement
parameters, and a value of 0.9 will make use of more voxels than a value of
0.75. As is often the case, there is an accuracy/speed tradeoff. Higher
values for this setting should result in higher accuracy - but a slower
algorithm. I figured that because computers are generally getting faster, it
shouldn't do any harm to adjust the settings slightly - hopefully resulting
in a slight accuracy increase.
>
> Is there anybody to answer this question?
I'm here for much of the pre-processing stuff (as long as the answer's are not
already in the mailing list archives multiple times).
Best regards,
-John
|