Adam Grant Chandler PhD, Presenter: Employee, General Electric Company

Tinsu Pan PhD, Abstract Co-Author: Nothing to Disclose

Delise Herron, Abstract Co-Author: Nothing to Disclose

Ella Anderson, Abstract Co-Author: Nothing to Disclose

Wei Wei, Abstract Co-Author: Nothing to Disclose

Chaan Ng MD, Abstract Co-Author: Nothing to Disclose

To compare the relative performance of manual, rigid, and non-rigid registration techniques for motion correction of lung tumors in breath hold CT images.

25 patient datasets, each consisting of six limited sequential breath-hold helical volumes through an index lung lesion and one reference cine image obtained from a CT perfusion protocol (4x5mm row MDCT), were evaluated. Each of the six helical volumes were registered to the reference image using 1) manual visual registration (two observers in consensus), and 2) two automated intensity-based registration methods: a) rigid-translational and b) non-rigid free-form deformations (with B-splines). For each method, 150 registrations were performed. The performance of each of the three registration techniques for the tumor region was assessed by 1) two quantitative alignment metrics (percentage overlap and distance of center of mass (DCOM)) and 2) visual validation (four observers blindly scored the degree of alignment of each registration on a 5-point scale, ranging from <2% to >50% visual misalignment).

The manual method showed the worst performance on both quantitative evaluations: average overlap and DCOM (77.6% and 2.99mm, respectively), compared to the rigid (87.7% and 1.08mm) and non-rigid (91.8% and 0.41mm) methods (all p<0.0001). Of the automated methods, nonrigid showed superior performance compared to rigid by both quantitative evaluations (p<0.0001). Visual validation confirmed these findings: clinically acceptable alignment (judged to be <10% misalignment visually) in 68.5% (411/600 observations) for manual cases, compared to 99.8% (599/600 observations) (p<0.0001) for rigid, and 100% (600/600 observations) (p<0.0001) for non-rigid cases. There was no statistical difference in clinically acceptable performance between the two automated methods (p>0.99).

Automated registration techniques achieve significantly better alignment than manual registration methods, with marginal superiority of non-rigid, compared to the rigid method. The latter, however, utilizes substantially less computational resources and may be just as acceptable clinically.

Motion correction is an important component of many imaging applications. It is a potentially important component in body CT perfusion studies, where motion can have detrimental effects.

Chandler, A, Pan, T, Herron, D, Anderson, E, Wei, W, Ng, C, Evaluation of Automated Registration Techniques for Motion Correction of Lung Tumors in Helical Breath Hold CT Images.  Radiological Society of North America 2008 Scientific Assembly and Annual Meeting, February 18 - February 20, 2008 ,Chicago IL. http://archive.rsna.org/2008/6017758.html