Level set segmentation extracts only one vessel at a time

as in this tutorial https://www.youtube.com/watch?v=DJ2032yo5Co
i try to use level set segmentation (after successful vesselness filtering) but the newer version of level set segmentation seems to only extract one vessel at a time based on a starting and ending seed points, I can’t get it to work with only one seed ( to extract the whole tree like in the tutorial). thanks for your help

Maybe the level set module used a different method a very many years ago, but it was later changed to the current method.

The old method looks similar to Fast marching. You can install SegmentEditorExtraEffects extension and try Fast marching, Grow from seeds, and Watershed effects in the Segment Editor module.

1 Like

it indeed looks like a basic extraction method, i was hoping it would take in account contours or any tree/ branch detection algorithm, thanks for your reply!

VMTK extension has centerline extraction and tree model builder module, too.

Thanks, i cant find the " tree builder " module anywhere, only have vesselness filtering, level set segmentation (wich does what i want but just for one at a time) and centerline extraction (wich uses an already made vascular model). i would just like to use vesselness output to segment the whole tree with a "tubular/branch seeking " algorithm instead of just painting it with threshold/ region growing (i assume the output would be a lot more refined)

By tree builder I meant centerline extraction module. It can work on the model exported from segmentation (in Data module, right-click on segmentation node and choose to export to model).

In the current workflow, branch searching happens after segmentation, on the generated model node. Why would you like to change the order? Are you aware of a “tubular/branch seeking” vessel segmentation algorithm in VMTK?

The segmentation step is the problematic one here, vesselness enhances the volume but its not perfect, using any general purpose tool (grow from seeds/threshold paint/ region growing) can yield satisfying results but often has artifacts or irregular diameters
No i did not see such a thing anywhere, just assumed it was part of Level set segmentation since it worked without artifacts in the video (but i now understand it only used something like region growing)

the purpose of this would be to “guide” the segmentation or model creation based on a tree-seeking logic (like centerline’s shape) thus keeping diameters as the average of the neighboring section, producing a more cylindrical shape and clean intersections. inputs would be either vesselness volume, a starting seed and a direction seed (materialise mimics did something similar for airway segmentation https://www.youtube.com/watch?v=OYncaVGtGEM )

for a practical example, here is a portal vein reconstruction after vesselness, after some smoothing , then i extracted the centerline and generated a tree model using it (this is just to show the tree generation concept, of course i’m talking about extraction right from a volume, without making an initial segmentation, thus taking in account the real diameter and shape data)

thanks for your answers, i hope i described the concept adequately

Hi, i came across this study which had me remember the branch based algorythm i spoke about, this uses the centerline tree to extrude varying diameter branches and output a clean model. tought this might be interesting.