# Manual segmentation slow to complete a contour

**URL:** <https://discourse.slicer.org/t/manual-segmentation-slow-to-complete-a-contour/9527>\
**Category:** Support\
**Tags:** segmentation, dicom\
**Created:** [December 17, 2019, 3:50pm UTC](https://discourse.slicer.org/t/manual-segmentation-slow-to-complete-a-contour/9527 "2019-12-17T15:50:14Z")\
**Posts on this page:** 1\
**Showing post:** 4

<div class="post-metadata">

**Author:** ![lassoan](https://sea2.discourse-cdn.com/flex002/user_avatar/discourse.slicer.org/lassoan/32/13_2.png) [@lassoan](https://discourse.slicer.org/u/lassoan)\
**Post date:** [December 18, 2019, 9:14pm UTC](https://discourse.slicer.org/t/manual-segmentation-slow-to-complete-a-contour/9527/4 "2019-12-18T21:14:06Z")

</div>

> [@spaper](#):
>
> The gaps in slice thickness compared to the in plane resolution? We are completing our segmentations using the plane with high resolution and interpolating between every other slice along the poorer resolution slice thickness.

Yes, I mean that you cannot use a simple interpolation to recover details that are missing between slices because interpolation can only reconstruct the original signal if the Nyquist criterion is fulfilled (your sampling is sufficiently dense compared to the spatial frequency of the signal). If you apply anti-aliasing filter to ensure the criterion is not violated, then the filtering removes the high-frequency content, so essentially the high-resolution details in the ultrasound slices cannot contribute to the 3D image reconstruction.

A common, very visible result of Nyquist criterion violation is staircase artifacts on the reconstructed surface. See [this](https://discourse.slicer.org/t/combining-volumes-what-am-i-missing/2941/2) and related posts for details.

You may be able to utilize the high in-plane resolution of the ultrasound images by using superresolution techniques, but not by simple image interpolation. If you don’t have time to explore such methods then I would recommend to remove the information content that you cannot use anyway, early in the data processing pipeline (downsample the ultrasound images), to avoid pushing through huge amount of data through the pipeline that actually cannot contribute to the end results.

> [@spaper](#):
>
> It is strange that with the same voxel sizes, the ultrasound images do not take this long to complete a contour on a single slice. It is only happening with the MRI images.

It is hard to guess what could go wrong. Maybe voxels in you MRI image are floating-point? If you can share a scene that demonstrates the unexpected performance issue or you can reproduce the problem with any of the public sample data sets (maybe resampling them using Crop volume module) then we can investigate this further.

---

_[View the full topic](https://discourse.slicer.org/t/manual-segmentation-slow-to-complete-a-contour/9527)._