Program Information
Evaluation of the Performance of a New Deformable Image Registration Algorithms Based On Motion Modeling for Cone-Beam CT Images
N Alsbou1*, S Ahmad2 , I Ali2 , (1) University of Central Oklahoma, Edmond, OK, (2) University of Oklahoma Health Sciences, Oklahoma City, OK
Presentations
WE-RAM2-GePD-J(A)-4 (Wednesday, August 2, 2017) 10:00 AM - 10:30 AM Room: Joint Imaging-Therapy ePoster Lounge - A
Purpose: To develop a deformable image registration (DIR) algorithm based on motion modeling that works as a standalone algorithm or provides guidance for different DIR-algorithms.
Methods: An algorithm was developed that performed DIR based on a motion model. This algorithm used different motion patterns to perform iterative deformation of the images to produce the shapes and patterns in the reference images. The performance of this DIR-algorithm was compared with four regular DIR-algorithms: Demons, Fast-Demons, Horn-Schunck and Lucas-Kanade from the DIRART software. This algorithm was tested with the CBCT-images of a mobile phantom scanned with a kV-on-board imager mounted on a Varian TrueBeam machine that was registered to the reference stationary CBCT-images. This phantom included three mobile targets made from water-equivalent materials inserted in low-density-foam which moved a cyclic motion with adjustable motion amplitudes and frequencies.
Results: The new algorithm is superior to regular DIR-algorithms where it produced better matching between the images of the mobile phantom and the stationary images. This algorithm solved the issues associated with large motion amplitudes where the regular DIR-algorithms failed to deform the mobile targets to produce the characteristics of the actual targets in terms of shape, size and CT-numbers even at large motion amplitudes. In CBCT-images, the distortion in the images intensity is mainly dependent on motion amplitude for cyclic motions which can be predicted easily with this algorithm. The performance of this algorithm can be improved significantly with prior knowledge of the patient or phantom motion that can be obtained, for example, from the tracking of an external or internal marker.
Conclusion: This modified DIR-algorithm considering motion modeling reproduced well the lengths and CT-numbers of the mobile targets. In contrast, the other DIR-algorithms based on the solution of the optical-flow-equation and attraction-forces failed in deforming the images particularly at large motion amplitudes.
Contact Email: