Accuracy of Automated Centerline Approximation Algorithms for Lower Extremity Vessels in CTA Phantom

Information

  • Publication Type: Conference Paper
  • Workgroup(s)/Project(s):
  • Date: March 2004
  • Booktitle: European Congres of Radiology
  • Location: Vienna, Austria
  • Organization: Institute of Computer Graphics and Algorithms, Vienna University of Technology
  • Keywords: Medical Visualization, Vessel Segmentation, Centerline Detection

Abstract

Purpose: The accurate determination of the central vessel axis is a prerequisite for automated visualization (curved planar reformation) and quantitation. The purpose of this work was to assess the accuracy of different algorithms for automated centerline detection in a phantom simulating the peripheral arterial tree. Methods and Material: Six algorithms were used to determine the centerline of a synthetic peripheral arterial vessel (aorto-to-pedal arteries, diameter 18-0.6mm) dataset (256x256x600, voxel size 0.5x0.5x0.5mm). They are ray-casting/thresholding (RCT), ray-casting/maximum gradient (RCMG), block matching (BM), fitting to ellipse (FE), center of gravity (CoG), and Randomized Hough transform (RHT). Gaussian noise whith a sigma: 0, 5 and 10 was used to observe the accuracy of the method under noise influence The accuracy of automatic centerline determination was quantified by measuring the error-distance between the derived centerlines, and the known centerline course of the synthetic dataset. Results: BM demonstrated unacceptable performance in large vessels (>5mm) when the shift used was less than 3 voxels. RCMG demonstrated a greater error (mean of the error 4.73mm) in large diameter (>15mm) vessels than in small diameter (<15mm) vessels (mean of the error 0.64mm). Because RHT and FE use Canny edge detector preprocessing, both are sensitive to noise. CoG and RCT keep the mean of the error-distance significantly smaller (0.7mm and 0.9mm respectively) than all other algorithms. Conclusion: CoG and RCT algorithms provide the most efficient centerline approximation over a wide range of vessel diameters.

Additional Files and Images

No additional files or images.

Weblinks

No further information available.

BibTeX

@inproceedings{alacruzECR2004,
  title =      "Accuracy of Automated Centerline Approximation Algorithms
               for Lower Extremity Vessels in CTA Phantom",
  author =     "Alexandra La Cruz and Mat\'{u}s Straka and Arnold K\"{o}chl
               and Milos Sr\'{a}mek and Meister Eduard Gr\"{o}ller and
               Dominik Fleischmann",
  year =       "2004",
  abstract =   "Purpose: The accurate determination of the central vessel
               axis is a prerequisite for automated visualization (curved
               planar reformation) and quantitation. The purpose of this
               work was to assess the accuracy of different algorithms for
               automated centerline detection in a phantom simulating the
               peripheral arterial tree. Methods and Material: Six
               algorithms were used to determine the centerline of a
               synthetic peripheral arterial vessel (aorto-to-pedal
               arteries, diameter 18-0.6mm) dataset (256x256x600, voxel
               size 0.5x0.5x0.5mm). They are ray-casting/thresholding
               (RCT), ray-casting/maximum gradient (RCMG), block matching
               (BM), fitting to ellipse (FE), center of gravity (CoG), and
               Randomized Hough transform (RHT). Gaussian noise whith a
               sigma: 0, 5 and 10 was used to observe the accuracy of the
               method under noise influence The accuracy of automatic
               centerline determination was quantified by measuring the
               error-distance between the derived centerlines, and the
               known centerline course of the synthetic dataset. Results:
               BM demonstrated unacceptable performance in large vessels
               (>5mm) when the shift used was less than 3 voxels. RCMG
               demonstrated a greater error (mean of the error 4.73mm) in
               large diameter (>15mm) vessels than in small diameter
               (<15mm) vessels (mean of the error 0.64mm). Because RHT and
               FE use Canny edge detector preprocessing, both are sensitive
               to noise. CoG and RCT keep the mean of the error-distance
               significantly smaller (0.7mm and 0.9mm respectively) than
               all other algorithms. Conclusion: CoG and RCT algorithms
               provide the most efficient centerline approximation over a
               wide range of vessel diameters. ",
  month =      mar,
  booktitle =  "European Congres of Radiology",
  location =   "Vienna, Austria",
  organization = "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology",
  keywords =   "Medical Visualization, Vessel Segmentation, Centerline
               Detection",
  URL =        "/research/publications/2004/alacruzECR2004/",
}