# Extract Centerline computational troubleshooting

**URL:** <https://discourse.slicer.org/t/extract-centerline-computational-troubleshooting/45322>\
**Category:** Support\
**Created:** [December 2, 2025, 2:51pm UTC](https://discourse.slicer.org/t/extract-centerline-computational-troubleshooting/45322 "2025-12-02T14:51:48Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Stephanie\_Baumgart](https://sea2.discourse-cdn.com/flex002/user_avatar/discourse.slicer.org/stephanie_baumgart/32/80723_2.png) [@Stephanie\_Baumgart](https://discourse.slicer.org/u/Stephanie_Baumgart)\
**Post date:** [December 2, 2025, 2:51pm UTC](https://discourse.slicer.org/t/extract-centerline-computational-troubleshooting/45322/1 "2025-12-02T14:51:48Z")

</div>

I was trying to use Extract Centerline to generate a centerline tree of what I think is a relatively simple shape (compared to what I want to input later). I’ve got 128gb RAM, 8GB NVIDIA card, 2TB SSD, and Intel i7-14700HX processor, which I would think would be enough to process at least this.

> **[Dropbox](https://www.dropbox.com/scl/fo/6avpncksx1dg0ex7vf6pn/ABLxyLXTrRmDcGauzTXJL_8?rlkey=pliyhrqs1loqqiqdaod92rpzf&st=hkv7ip67&dl=0)**

The model was imported as a .ply from an Avizo export because the lab I’m in segments in Avizo (and not going to change). It was about 42k points, I used decimate set at 0.9 and it output a 23k-point mesh.

If I did not run pre-processing, the network outputs rendered very quickly, though missing a couple branching points I had marked with endpoints, and the centerline outputs haven’t finished after 30min. If I did run pre-processing, Slicer stays pre-processing for over 30min. I’ve tried changing curve sampling distance to 0.001mm because the scan resolution is 0.006mm; I saw that was an issue in a previous post. I’ve also tried increasing target point count. It pre-processes fast when set to 23k (basically no more decimating), then spends tens of min on the centerline curve and table generation. When I set target point count for 15k, then it takes a very long time to pre-process again, I haven’t seen it finish either step yet.

If this was the most complicated tree structure I wanted a centerline tree of, then I’d be more patient. But I have much more complex structures that I will need trees from too, so I was wondering if there was anything else I could do to optimize computation time or if I was missing something obvious. Avizo has a similar module, but it finishes the computation in a minute or two, though the line isn’t always in the center and it’s harder to edit/refine (talking with them about that too), so I was hoping to use Slicer’s.

Thank you for any ideas!

---

<div class="post-metadata">

**Author:** ![chir.set](https://sea2.discourse-cdn.com/flex002/user_avatar/discourse.slicer.org/chir.set/32/66982_2.png) [@chir.set](https://discourse.slicer.org/u/chir.set)\
**Post date:** [December 2, 2025, 5:32pm UTC](https://discourse.slicer.org/t/extract-centerline-computational-troubleshooting/45322/2 "2025-12-02T17:32:29Z")

</div>

Your data has a `tree_smooth` and a `decimate tree` models.

With multiple combinations of `Preprocess input surface`, `Target point count` and `Subdivide`, a centerline model only (Tree) is generated in a few seconds with either input model.

Trouble starts while generating centerline curves, it runs and runs and there’s no console output hinting at anything.

If the bifurcated centerline (11K+ points) is processed with `vtkCleanPolyData` and a new model is created from its output (4K+ points), and the clean centerline model is processed in `Centerline disassembly`, its components as models are created in a few mins. But it keeps running when curve components are requested. `Centerline disassembly` relies on `Extract centerline` to generate curves.

If the bifurcated centerline itself is processed in `Centerline disassembly` (without cleaning), the same observations can be made.

So the lengthy computation happens with centerline curve generation in `Extract centerline` for unknown reasons. I’m afraid you’ll have to work with centerline models for now.

Someone else may have a better idea.