Frangi vesselness is an algorithm for detection of tube-like structures (such as in imagery of filamentous structures (blood vessels, neurites, etc.). It applies to both 2D and 3D images and was first described by Frangi et al 1998.1 (PDF). In many cases, this method is known to be a better alternative to single-scale Tubeness filtering (at least for isotropic images), but it is slower. There are two ImageJ implementations of the algorithm:
-
A legacy plugin (In Fiji registered under Plugins › Process › Frangi Vesselness (imglib, experimental), source code here). This is a deprecated implementaton with expected inaccuracies.
-
An ImageJ Op (In Fiji registered under Process › Filters › Frangi Vesselness). This version supersedes previous implementations and has been validated/benchmarked againgst other implementations (namely ITK and MATLAB) (details). It requires 3 inputs:
- Scale(s): Radii of the structures to be filtered
- Spacing: the physical spacing between image data points (i.e., the pixel/voxel dimensions)
- Whether or not a gaussian filter should be applied at each scale before the filter runs
SNT’s uses this filter internally. Its Secondary Layer Wizard provides a convenient way to preview the effect of scale size in the result.
Frangi Vesselness: Left: Original image. Right: Filtered image.
