Hi @Sverre & @cnet ,
sure, PET images look like this, especially if you look at them with a software which does not interpolate between pixels. These days, PET scanners don’t exist alone, usually combined PET-CT scanners are available in the hospitals. Furthermore, PET images are usually taken together with CT images (at least for attenuation correction). Physicians draw contours (e.g. for therapy planning) usually on the CT images with the PET images fused. In typical clinical software, both PET and CT images are shown with a trilinear interpolation. But finally, contours are saved in the coordinate system of the CT images (at least in most radiotherapy treatment planning software I have seen).
To make a long story short: What about upsampling the images (ImageJ Menu Image > Scale) with tri-linear interpolation to fit the voxel size of the corresponding CT image stack (usually something like 1.5x1.5x5 mm)? That does not eliminate the issue of being not precise, but PET scanners do have a “bad” resolution, so there may be no easy way to deal with that otherwise…
BTW: If you use an automatic thresholding approach to measure the volume of the area with high activity (PET signal), it may result in volume measurements which are mostly independent of if you use the liniear interpolated images or the original. The contours just look nicer in the interpolated case