Encrypted login | home

Program Information

McSART: An Iteratively Updated Motion-Model and Image Reconstruction Algorithm

no image available
G Chee

G Chee*, D Low , D O'Connell , K Singhrao , J Lewis , University of California, Los Angeles, Los Angeles, AA


WE-F-605-5 (Wednesday, August 2, 2017) 1:45 PM - 3:45 PM Room: 605

Purpose: To create a respiratory motion model from CBCT projections by iteratively alternating between updating the motion model and performing a motion-compensated CBCT reconstruction.

Methods: A digital eXternal Cardiac Torso (XCAT) phantom was created with a 5s regular breathing period and 2cm maximum diaphragm motion. CBCT image acquisition was simulated with 720 projections obtained every 0.5°. The projections were binned into 6 phase bins for 4DCBCT reconstruction with a simultaneous algebraic reconstruction technique (SART). These bins were used to generate a time-independent respiratory motion model that estimates lung motion based on patient diaphragm position. This motion model, alongside known diaphragm displacements, was used to reconstruct images at a reference frame within each bin in a motion-compensated SART (McSART) algorithm. The resulting images were used to update the motion model, which is used in the following McSART iteration. Each iteration increases the number of projections included in each bin, under the assumption that the motion model can correctly compensate for small motions outside the bin.

Results: The iterative updates improved the accuracy of the motion model. Sum-squared differences were calculated between the model-predicted DVFs at the 1st and 15th update and XCAT-derived DVFs. These decreased by a factor of 20 in the AP and LM directions and a factor of 4 in the SI direction, with the SI direction giving the smallest differences. Normalized pixel value differences between the 1st and 15th McSART images and ideal SART images (assuming a motionless phantom) improved by a factor of 2 over a region-of-interest around the diaphragm.

Conclusion: The iterative motion model is able to predict more accurate images than 4DCBCT images. The image reconstruction in McSART is promising but requires further tuning. Future work will optimize the image reconstruction parameters, and shift towards using real patient CBCT projections and breathing patterns.

Funding Support, Disclosures, and Conflict of Interest: Research funded by Varian Medical Systems

Contact Email: