Report from SCBT/MR: Vascular centerlines come clear on peripheral occlusion CTA

April 3, 2006

A new image processing software program skirts some of the pitfalls inherent with current systems, according to a study presented this week in Phoenix at the annual meeting of the Society of Computed Body Tomography and Magnetic Resonance.

A new image processing software program skirts some of the pitfalls inherent with current systems, according to a study presented this week in Phoenix at the annual meeting of the Society of Computed Body Tomography and Magnetic Resonance.

Volume-rendered images and maximum imaging projections have their place in CT image interpretation, but they fall short when it comes to assessing severely diseased peripheral arterial occlusive vessels, said lead author Dr. Justus Roos, a radiology fellow at Stanford University.

Curved planar reformations deliver the most information about occlusion morphology and length, but current methods for generating CPR have shortcomings. Overcoming these limitations is the focus of an interdisciplinary Stanford team's efforts.

CPR generation requires the extraction of a vascular centerline. This can be done manually, but it's a time-consuming process, particularly with long peripheral occlusions. Or it can be done automatically.

Available extraction software relies on density and gradient information. These algorithms work well in normal or mildly diseased vessels, but they consistently fail in severely diseased and occluded segments.

"The density- and gradient-based algorithms fail because the density of diseased arteries is simply very different - higher or lower - from normally perfused vessels. So currently available software does not work in those segments that are actually diseased, which are probably why a patient was scanned in the first place," Roos told Diagnostic Imaging.

The researchers decided to take advantage of the similarity of vascular shapes in different individuals. They sought to find a way to extract this "high order shape information" to find centerlines through occlusions independent of density and gradient information.

They built their knowledge-based algorithm from CT angiography femoropopliteal data sets acquired from a combination of normal (30) and diseased (20) patients. They first identified the arteries using a standard density-based semiautomated vessel tracking algorithm. Occluded segments were manually traced two times by two experts, with the average of all four readings serving as the reference centerline.

Each occlusion was also tracked using a knowledge-based algorithm, which builds on prior knowledge of shape information. The new algorithm was tested and compared with the reference standard.

The knowledge-based algorithm successfully tracked all femoropopliteal occlusions. The longer the occlusion, however, the more the automated centerline deviated from the experts' manually derived centerline:

  • 0.1 mm average deviation for occlusions less than 75 mm

  • 3.7 mm deviation for occlusions 75 mm to 150 mm

  • 16.7 mm deviation for occlusions greater than 150 mm

These are the first results of a new shape-based approach to centerline identification. This stage is a proof of concept, but the results are encouraging enough for the researchers to further pursue this strategy, Roos said.

"We cannot predict how accurate the method might become in the future, but traditional algorithms that exploit the density information of the opacified vessels seem to be a dead end," he said.

Stanford's interdisciplinary research strategy is somewhat unusual, according to Roos. Researchers develop new algorithms in a very close clinical environment using clinical cases early on. They implement new algorithms immediately so that clinical radiologists can get a feel for it.

"This shortcuts inefficient long feedback loops," he said.

While researchers refine the knowledge-based algorithm for femoropopliteal occlusions, they also are developing a shape-based approach for ileac arteries, whose tortuosity makes the task more challenging, Roos said.

For more information from the Diagnostic Imaging archives:

Radiologists seek more control over 3D postprocessing

Dedicated labs lend consistency to image quality and results

MSCT angio advances in peripheral vessels

IR popliteal aneurysm fix moves to center stage