Encrypted login | home

Program Information

Breathing Motion Model Estimation From Deformable Registration of CBCT Projections

no image available
D O'Connell

D O'Connell1*, G Chee1 , J Lewis1 , D Thomas2 , L Yang1 , P Lee1 , D Low1 , (1) University of California, Los Angeles, Los Angeles, CA, (2) University of Colorado, Denver, CO


MO-F-205-5 (Monday, July 31, 2017) 4:30 PM - 6:00 PM Room: 205

Purpose: Cone-beam scanning devices (CBCT) mounted on linear accelerators are susceptible to breathing motion artifacts in reconstructed images due to the slow gantry rotation speed. In this work, we propose a novel method to estimate a motion model directly from projection data using 2D deformable image registration (DIR). The motion model can be used to generate respiratory correlated images (4DCBCT) without binning projections.

Methods: CBCT acquisition was simulated using a patient CT image and GPU-accelerated forward projector. Respiratory motion was approximated by deforming the image volume from end-exhalation to full inspiration according to a pre-existing motion model as projections were calculated. An initial reference image was provided. At each gantry angle, a projection through the static reference image was compared to a projection through the dynamic volume using DIR. The 2D deformation vector fields (DVF) were backprojected and used to iteratively update estimates of voxel-specific parameters for a linear motion model. The reference image was deformed to full inspiration using the motion model obtained with the proposed method and compared to the ground-truth image by using DIR and measuring the magnitude of deformation vectors.

Results: The full inspiration image reconstructed using the estimated motion model agreed with the ground-truth image to 1.26 +/- 1.1 mm on average. In comparison, the 95th percentile tissue motion during acquisition was 18.1 mm. Blurring at the diaphragm was markedly reduced compared to a reconstruction performed without motion compensation.

Conclusion: A method to estimate respiratory motion model parameters from projections was developed and tested on a digital phantom. The method is straight-forward and flexible in that it can be used with any 2D DIR algorithm. The proposed technique can be integrated into an alternating scheme to estimate the motion model parameters and reference image simultaneously.

Funding Support, Disclosures, and Conflict of Interest: Dr. Low reports grants from Siemens Healthcare and Varian Medical during the conduct of the study.

Contact Email: