Premium
Comparing Performance of Centerline Algorithms for Quantitative Assessment of Brain Vascular Anatomy
Author(s) -
Diedrich Karl T.,
Roberts John A.,
Schmidt Richard H.,
Parker Dennis L.
Publication year - 2012
Publication title -
the anatomical record: advances in integrative anatomy and evolutionary biology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.678
H-Index - 62
eISSN - 1932-8494
pISSN - 1932-8486
DOI - 10.1002/ar.22603
Subject(s) - tortuosity , segmentation , computer science , algorithm , ground truth , artificial intelligence , mathematics , computer vision , geology , geotechnical engineering , porosity
Attributes like length, diameter, and tortuosity of tubular anatomical structures such as blood vessels in medical images can be measured from centerlines. This study develops methods for comparing the accuracy and stability of centerline algorithms. Sample data included numeric phantoms simulating arteries and clinical human brain artery images. Centerlines were calculated from segmented phantoms and arteries with shortest paths centerline algorithms developed with different cost functions. The cost functions were the inverse modified distance from edge (MDFE i ), the center of mass (COM), the binary‐thinned (BT)‐MDFE i , and the BT‐COM. The accuracy of the centerline algorithms were measured by the root mean square error from known centerlines of phantoms. The stability of the centerlines was measured by starting the centerline tree from different points and measuring the differences between trees. The accuracy and stability of the centerlines were visualized by overlaying centerlines on vasculature images. The BT‐COM cost function centerline was the most stable in numeric phantoms and human brain arteries. The MDFE i ‐based centerline was most accurate in the numeric phantoms. The COM‐based centerline correctly handled the “kissing” artery in 16 of 16 arteries in eight subjects whereas the BT‐COM was correct in 10 of 16 and MDFE i was correct in 6 of 16. The COM‐based centerline algorithm was selected for future use based on the ability to handle arteries where the initial binary vessels segmentation exhibits closed loops. The selected COM centerline was found to measure numerical phantoms to within 2% of the known length. Anat Rec, 2012. © 2012 Wiley Periodicals, Inc.