OutOfMemoryError: CUDA out of memory TotalSegmentator

Operating system: Windows 10
Slicer version: 5.2.2
Expected behavior: TotalSegmentator full resolution segmentation with GPU 8Gb (NVIDIA)
Actual behavior: semgmentation fails due to “OutOfMemoryError: CUDA out of memory”, seems to result from preprocessing crop of CT images (509 in that case). Lower croping makes the process succeed (around 250).
How does it work and is it possible to edit this parameter in TotalSegmentator.
One 16 Gb GPU is best but we can read that 8Gb is also sufficient (our IT dept did not offer the first one…).
Thank you for your help,

Does it work with the CTs from the Sample datasets?

Hi Rudolf,

Yes ! thanks for the tip.

As consequence, the resolution is lowered with this CT (2.2 vs 0.9 mm). With the high resolution of segmentation (1.5mm vs 3mm), it could be worth being quantified, nevertheless.

Anyway, we are not blocked anymore and that is the most important, I would say!

Nevertheless, do you have any clue on how the resampling operate within TotalSegmentator?

Thank a lot for your help and answering,

Best regards,


You can crop and/or resample the volume on the CPU using Crop volume module before you run Total Segmentator on it.

Thanks Andras,

It works pretty well and ensure to perform TotalSegmentator on the high-resolution CT (relatively).

And this remains a quick pre-processing step.

Best regards,


Concerning TotalSegmentator, just a quick wondering, any idea why the sternum is not included?



There is a preview circulating with sternum already, it will probably soon be integrated.
Best, r

The latest TotalSegmentator model includes the sternum, but I don’t think it has been publicly released yet.