Centerline Extraction

[lisacptnhook]

Hello, I am thinking about integrating the centerline extraction in my Software. But, unfortunately, I am not able to produce any centerlines. I tried different examples, e.g. this one here
VMD_Aorta_0030.zip
which produces sensible distance image and centerline candidates, but the Fast Marching Algorithm doesn't work. I'd be very thankful for any hints or help :)

I checked out the C++ repository and compiled from source under Ubuntu. I use this code:

auto importer = ImageFileImporter::New();
 importer->setFilename("data/VMD_Aorta_0030.mhd");

auto centerline = CenterlineExtraction::New();
centerline->enableRuntimeMeasurements();
centerline->setInputConnection(importer->getOutputPort());
    
// Export mesh to disk
auto exporter = VTKMeshFileExporter::New();
exporter->setInputConnection(centerline->getOutputPort());
exporter->setFilename("data/centerline.vtk");
exporter->setWriteNormals(true);
exporter->update();

[smistad]

Hi

There were two issues with the volume which the centerline extraction implementation wasn't able to handle.
The fact that the segmentation volume is stored using data type short (int16), and that the segmentation goes to the edge, it needs some padding.

I quickly tested converting the segmentation volume and extracting the centerline and visualizing it with FAST in Python. That worked and produced a nice centerline of the aorta:

import fast
import numpy as np

importer = fast.ImageFileImporter.create('VMD_Aorta_0030.mhd')
image = importer.runAndGetOutputData()

data = np.asarray(image)[..., 0] # Convert to ndarray and remove channel dimension
data = data.astype(np.uint8) # Convert to uint8
data = np.pad(data, 12) # Pad with zeros

image2 = fast.Image.createFromArray(data)

centerline = fast.CenterlineExtraction.create().connect(image2)

renderer = fast.LineRenderer.create().connect(centerline)

fast.SimpleWindow3D.create().connect(renderer).run()

I admit the centerline extraction method should have handled these two issues:

1. Be able to process data types other than uint8, or at least produce an error message, and
2. Be able to handle segmentations at the edges of the volume.

[lisacptnhook]

Thank you so much for the fast reply! With conversion and padding I was able to extract the centerlines!

[lisacptnhook]

I have a short follow up question: the algorithm seems to modify the coordinate system. If I load the computed centerline together with the original label (as surface mesh) the centerline does not lie within the surface mesh. Do you have any hints for me as how the coordinates are modified, so that I can extract the centerline within the original coordinate system?

[smistad]

I assume you mean that in the mhd file there is a transform which applies for the volume in question.
FAST keeps such transforms in a scene graph, and the transform should be applied when you use the VTKMeshFileExporter.
Most process objects preserve the scene graph by adding new output data objects as child of the input data object.
E.g. in C++: SceneGraph::setParentNode(output, input);
However, if you have some custom process objects in your processing chains, you need to do this by yourself.
The quick python example above breaks the scene graph, because it has an intermediate numpy processing.