Accuracy of segmentation smoothing

[lassoan (Andras Lasso)]

@pieper

Also, these CT pixels are about .5 mm so the differences appear in most cases to be about .1 or .2 mm or less. It would be interesting to hear your thoughts on the accuracy you need to achieve.

Yes, the the segmentation results look about the same to me, with the difference being smaller than a single voxel.

See intensity profile along a line orthogonal to the bone surface. The width of the transition zone from soft tissue to bone is about 2mm. It would be hard to tell where is the exact boundary: at the peak intensity, where the peak starts, at half maximum, …? Each definition could be valid and the difference between them can be up to a millimeter.

@vikas26 What is the accuracy requirement for your clinical application?

@mikebind

More conversion settings (such as a decimation factor) are accessible in the Segmentations module, if you click the “Update” button next to the “Closed Surface” line in the “Representations” section, and then click the ‘Binary labelmap → Closed surface’ path in the window that pops up.

I would specifically recommend to try the surfacenet smoothing filter:

• Conversion method: 1 (vtkSurfaceNets3D)
• SurfaceNets smoothing: 1 (smoothing done in surfacenets filter)