Purpose: A free-breathing (FB) CBCT could suffer from respiratory motion-induced artifacts due to the limited gantry-rotation speed. This study sought to develop an iterative image-to-image (III) reconstruction algorithm to remove motion-induced artifacts in FB-CBCT.
Methods: The III-reconstruction algorithm starts with using a 4DCT to develop a deformable respiration model, which is subsequently used to map a presumed reference image to generate a 3D image for each of the respiration phases. The resultant phase CTs were weighted to create an average CT. By comparing the average image with the blurred FB-CBCT, a set of composition equations was established and inversely solved using a maximum-likelihood estimation-maximization (MLEM) algorithm. The algorithm was validated using 4DCT, 4D-CBCT and FB-CBCT of a cylindrical phantom which can be operated in static or moving modes. The entire process was demonstrated based on the 4DCT and FB-CBCT data of a lung cancer patient. Three metrics, universal-quality-index (UQI), signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR), were used to evaluate the quality of the reconstructed images.
Results: Compared to the CBCT of the static phantom, the CBCT of the moving phantom had UQI, SNR and CNR equal to 0.92, 1.86 and 8.1. After applying the III algorithm, these metrics increased to 0.95, 2.4 and 23.4, respectively. The III-reconstructed images outperformed the acquired 4D-CBCT by 0.92 to 0.85 for UQI, and 2.13 to 1.43 for SNR. For the FB-CBCT, scanned 4D-CBCT and III-reconstructed CBCT, their CNRs are 0.83, 20.86 and 14.72, respectively. For the lung dataset, the III-reconstructed image outperformed the clinical 4D-CBCT with SNR increased from 1.74 to 2.27, and UQI increased from 0.90 to 0.97.
Conclusion: A novel image reconstruction method has been developed to remove motion artifacts in FB-CBCT. This method does not require 4D-tracking devices for CBCT data acquisition and could be widely used in routine clinic.