Abdominal Aortic Aneurysms:

Three-dimensional Imaging of Abdominal Aortic Aneurysms using Centerline-based Analysis: Technique and Examples

Kevin M. Baskin, MD, Elvira V. Lang, MD, Shereen Chang, MS, William Stanford, MD, and Eric A. Hoffman, PhD



I. Introduction.

A. Problem: Human abdominal aortic aneurysms are often both tortuous and eccentric - properties which may substantially corrupt their accurate measurement from conventional two-dimensional (transverse) computed tomographic images. Two examples are provided below which illustrate such potential errors.
1. Case 1. 76 year old male with a tortuous abdominal aortic aneurysm.
2. Case 2. 53 year old male with a primarily eccentric abdominal aortic aneurysm.
3.Measurements of aneurysm tortuosity and eccentricity from Volumetric Structural Analysis (VSA) of 46 human abdominal aortic aneurysms.
B. Impact of measurement of AAAs on patient outcome:
1. AAAs affect 3-4% of all Americans.
2. The outcome of a ruptured AAA is fatal in 78-94% of cases.
3. Repair should be undertaken when risk < risk of rupture.
4. Widespread disagreement re: risk and timing of repair.
5. Disagreement stems primarily from uncertainty.
C. The case for a relationship between diameter and AAA rupture:
1. Direct measurement of AAAs: cadaveric specimens (Darling, et al.)
2. Indirect measurement of AAAs:
a. Two-dimensional modalities: angiography, CT, US, MRI
b. Three-dimensional reconstructions: CTA, MRA, 3D-US
3. Needed: a reliable external standard.
D. Purpose: to develop a valid, reliable, and generalizeable approach to the measurement of AAAs, as a model anatomic structure with complex three-dimensional morphology.

II. Imaging Techniques

A. The accuracy of AAA measurements was tested with phantom aneurysms constructed of hand-blown Plexiglas, intended to span clinically relevant range.
B. Imaging protocol:
1. Helical CT scanner
a. 1:1 pitch
b. 320 mm field of view
c. Tube current = 100 mA, kVp = 120
d. Thin- and thick-slice collimations
e. Rotated through 3x2x2 matrix of rotations
i. transverse (0o, 45o, 90o)
ii. coronal (0o, 30o)
iii. sagittal (0o, 20o)
C. Imaging analysis and interpretation:
1. Conventional 2D-CT interpretation
a. Two radiologists (WS and BT) in a double-blind study
b. Maximum diameter to nearest mm, or
c. Diameter perpendicular to direction of tortuosity
d. Tortuosity scored on a three-point scale
2. Volumetric Structural Analysis (VSA)
a. Modular, task-oriented system for multidimensional image display and analysis
b. IMPROMPTU: semi-automated segmentation
c. Multi-Object Shape-based Interpolation: cubic voxels
d. Tube Geometry Analysis : greatest diameter
3. Analysis of tortuosity:
a. Tortuosity defined by a scalar angle, "alpha."
b. Represents the dot product of a unit vector coincident with the local centerline of the phantom, and a unit vector along the z-axis of the scanner,
c. alpha = arc-cos(vxhy-vyhx).
4. Analysis of eccentricity: defined as the aspect ratio at the greatest orthonormal diameter.

III. Our phantom studies demonstrated significant inaccuracies in conventional analysis:

A. Regression analysis.
1. Consistent underestimation of diameter by readers using conventional 2D-CT images, regardless of collimation.
a. Reader One: Thin-slice Protocol.
b. Reader Two: Thin-slice Protocol.
c. Reader One: Thick-slice Protocol.
d. Reader Two: Thick-slice Protocol.
2. VSA shows no such bias: correspondence of VSA with true dimension is much better.
a. Volumetric Structural Analysis: Thin-slice Protocol.
b. Volumetric Structural Analysis: Thick-slice Protocol.
B. Analysis of error data.
1. Human readers versus VSA.
a. The main effect of reader was highly significant (p < 0.0001).
b. Supplemental two-tailed t-tests (p = 0.01) demonstrated that reader 1 had less error than reader 2 (9.4% vs 12.5% error), and both readers had more error than VSA (0.4% error).
2. Effect of image collimation.
a. No overall impact of slice thickness on measurement accuracy was demonstrated by analysis of variance, (p = 0.4698).
b. a significant interaction was found between the error of measurements of individual readers and collimation (p = 0.0002).
3. Effect of phantom eccentricity.
a. Phantom eccentricity was found to have a very significant effect on error (p < .0001), with greater eccentricity resulting in more error.
b. There was also a significant interaction of eccentricity with reader (p < .0001), based on:
i. significant differences among the three levels of eccentricity for the human readers, and
ii. lack of increase in error between the two highest levels of eccentricity for VSA.
c. VSA also exhibited much less difference in error between the three levels of eccentricity than did the human readers.
4. Effect of phantom tortuosity - within the range studied,
a. Not found to significantly affect error
b. Not found to have differential effects for different readers.


IV. Discussion.

A. Radiologists' (2D-CT) measurements are precise, but biased and inaccurate.
B. Available corrective algorithms are too restrictive.
C. VSA is relatively immune to variations in 3D morphology.
D. A significant effect of tortuosity not demonstrated by this study.
E. Take-home points:
1. Natural history studies, outcomes analysis and management decisions for patients with AAAs have been based upon inaccurate and unstandardized measurements.
2. The magnitude of effect of these errors on patient management, outcome, and cost is unknown with respect to aneurysmal disease.
3. Given the availability of analytical tools (such as VSA) which are significantly more valid and reliable,
a. risk factors in the natural history of AAAs must be be reassessed, and
b. outcomes analysis of interventions must be reevaluated.








©1994-99 Division of Physiologic Imaging, Dept. of Radiology, Univ. of Iowa


Aneursyms | Tutorials | DPI Homepage | VIDA | NLM | Contact Us | Search



Last modified: Tue Jun 1 13:42:14 CDT