Segmentations Module: Exporting labelmaps using a transformed Reference volume

Hello, I have some question concerning translations and resampling of segmentation objects.

Background: I have two imaging volumes and a segmentation in my Data. Image volume (‘alpha’) is the original medical image, unaltered. Image volume (‘bravo’) is downsampled and translated to a slightly new origin. In addition to these volumes, I have a segmentation object. It’s unaltered by any transformations or resampling, and as such, the Master volume remains the original imaging volume.

Objective: To perform the same translation and downsampling on this segmentation, so it accurately maps to the image volume bravo.

Option 1) Convert to a binary labelmap, and perform resampling on the labelmap. In my particular use case, I am downsampling with interpolation, and specifying a new origin to create a translational shift. All parameters will match imaging volume bravo.

Option 2) In the Segmentations module I can directly export a binary labelmap with the Reference volume (in ‘advanced’) set to imaging volume bravo.

In both situations, I can verify the segmentation has been accurately translated and resampled in the Volumes module. My questions are:
A)In option 2 (exporting directly using reference volume bravo), what interpolation, if any, was performed to derive the new labelmap?
B) I noted that method 1 yielded a lower quality segmentation (before and after picture attached, after being the one with little unsegmented islands in it). Is option 2 a more standard approach than than option 1?

Thank you for your advice.

32%20PM 04%20PM

Nearest neighbor “interpolation” is used, i.e., there is no interpolation but the value of the closest input voxel is used.

There can be many explanations. One possible reason is that voxel values are stored in integer values (unsigned char), 0 = outside, 1 = inside. If you interpolate the value then due to numerical instabilities you might get 0.9999 as interpolated value, which may be converted to 0 integer value. If you want to interpolate (e.g., because you want to have smoother edges when you oversample) then probably you need to rescale values to 0=outside, 255=outside, apply Gaussian smoothing, resample with linear or spline kernel, then threshold with level=127.