Dear Keith,

>
> I'm upsampling time series data from 2.5mm to 2.0mm isotropic, using cubic interpolation.  The upsampled data seemed fine, but when I look at metrics like temporal stdev, I see some kind of grid-like artifact.  I tried sinc interpolation (basically the same as spline) and trilinear (MUCH bigger artifact).
>
> I'm attaching some representative screenshots.  For me, the more subtle artifact in the sinc+spline images is easier to see if you start with the trilinear.
>
> Is this expected?  Am I doing something wrong?

yes, I think it is kind of expected. Every type of interpolation results in a degree of smoothing/noise reduction. You have chosen to upsample in a way such that 4 original voxels comprise 5 new voxels. That means that if you start your grid on an “original voxel centre” (where you have no noise reduction) you will then sample your next voxel a little bit away from the original voxel centre -> some noise reduction, the next voxel midway between two original voxel centres -> maximum noise reduction, next some noise reduction, next no noise reduction etc etc.

Trilinear interpolation gives by far the worst smoothing effect, i.e. the largest noise reduction, which is why it is most obvious for that.

I hope I managed to make that clear.

Jesper


>
> Thanks!
> -Keith
>
>
> <upsamp2mm_sinc_tstd.png><upsamp2mm_trilinear_tstd.png><upsamp2mm_spline_tstd.png><orig25mm_tstd.png>