Four-dimensional cone-beam CT reconstruction based on motion-compensated robust principal component analysis

Abstract

To propose a motion compensation reconstruction method based on robust principal component analysis (RPCA) to reduce the influence of streak artifacts on accurate estimation of interphase motion deformation fields. We propose a RPCA motion compensation reconstruction algorithm to improve the estimation of motion deformation fields based on the traditional MC-FDK algorithm. RPCA was used to decompose the cone-beam computed tomography (CBCT) images into low-rank and sparse components, and the motion deformation fields between different phase images were then estimated using Horn and Schunck optical flow method from the low-rank images to reduce the influence of striping artifacts on the accuracy of estimation of interphase motion deformation fields. The performance of the algorithm was evaluated using simulation data and real data. The simulation phantom data was obtained by back-projection of 4D-CT images acquired from Philips 16-slice spiral CT using MATLAB software programming according to the scanning geometry of Varian Edge accelerator. The real patient data were obtained using the Elekta Synergy system of CBCT scanning system with half-fan mode CB projection data from lung cancer patients. Compared with images reconstructed using the traditional MC-FDK algorithm, the reconstructed image using the proposed method had clearer tissue boundaries with reduced motion artifact was reduced. The results of phantom data reconstruction showed that compared with the MC- FDK algorithm, the proposed algorithms resulted in improvements of PSNR by 25.4% and SSIM by 7.6%; compared with the FDK algorithm, PSNR was improved by 37.9% and SSIM by 17.6%. The proposed algorithm can achieve accurate estimation of inter-phase motion deformation fields and improve the quality of the reconstructed CBCT images.