Premium
Quantification of accuracy of the automated nonlinear image matching and anatomical labeling (ANIMAL) nonlinear registration algorithm for 4D CT images of lung
Author(s) -
Heath E.,
Collins D. L.,
Keall P. J.,
Dong L.,
Seuntjens J.
Publication year - 2007
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.2795824
Subject(s) - image registration , contouring , artificial intelligence , computer vision , landmark , nonlinear system , computer science , matching (statistics) , pattern recognition (psychology) , mathematics , nuclear medicine , image (mathematics) , medicine , statistics , physics , computer graphics (images) , quantum mechanics
The performance of the ANIMAL (Automated Nonlinear Image Matching and Anatomical Labeling) nonlinear registration algorithm for registration of thoracic 4D CT images was investigated. The algorithm was modified to minimize the incidence of deformation vector discontinuities that occur during the registration of lung images. Registrations were performed between the inhale and exhale phases for five patients. The registration accuracy was quantified by the cross‐correlation of transformed and target images and distance to agreement (DTA) measured based on anatomical landmarks and triangulated surfaces constructed from manual contours. On average, the vector DTA between transformed and target landmarks was 1.6 mm . Comparing transformed and target 3D triangulated surfaces derived from planning contours, the average target volume (GTV) center‐of‐mass shift was 2.0 mm and the 3D DTA was 1.6 mm . An average DTA of 1.8 mm was obtained for all planning structures. All DTA metrics were comparable to inter observer uncertainties established for landmark identification and manual contouring.