Program Information
A Novel 4D CBCT Reconstruction Method Using Registraition Assisted Compressed Sensing (RACS)
Z Qi1*, (1) University of Cincinnati, Cincinnati, OH
Presentations
TH-CD-303-8 (Thursday, July 16, 2015) 10:00 AM - 12:00 PM Room: 303
Purpose: Compressed Sensing (CS) based algorithms have demonstrated success in 4D Cone Beam CT by reconstructing high quality 4D images without significantly slowing down the acquisition. However, challenges remain, especially when data under-sampling is exacerbated by the use of a displaced detector; in such a scenario, the delineation of small pulmonary structures and their motion may suffer. The purpose of the study is to evaluate a novel algorithm named Registration Assisted Compressed Sensing (RACS), which is proposed to address the issue.
Methods: The RACS algorithm includes the following steps: (1) an initial 4D data set is reconstructed by a CS algorithm; (2) deformable registration is applied to calculate the displacement field between each phase and a reference phase; (3) The displacement fields are then used in a modified FBP algorithm to reconstruct for each phase a seed image, which is used to reconstruct the final image by CS. For validation, a numerical phantom study was conducted, in which a large sphere (2cm diameter) is moving simultaneously along with two small spheres (2mm diameter) placed on both sides, with a period of 4 seconds and an amplitude of 1cm. A 60-second cone beam CT scan was simulated. Both the CS and the RACS algorithms were applied for 4D reconstruction. In addition to visual comparison, line profiles through the centers of the all three spheres are compared between the two algorithms.
Results: The small spheres are clearly visualized in RACS reconstructed images, whereas they disappear in CS reconstructed images. The Root Mean Square Errors (RMSE) of the line profiles through the centers of the spheres are 11.2% and 4.3%, for the CS and RACS images, respectively.
Conclusion: By integrating deformable registration derived temporal information into reconstruction, the proposed RACS algorithm demonstrates a marked improvement in preserving small structures in reconstructed 4D images.
Contact Email: