Evaluation of Complimentary Rebinning Reconstruction for Off-Center Flat Panel Detector CBCT Imaging

Off-center flat panel detectors have been utilized in CBCT imaging to obtain a larger field of view such as the thoracic and upper abdomen regions. As a compromise, truncation artifacts arising from the asymmetrical geometry further degrades the image that is already commonly affected by motion-induced artifacts. Apart from the common iterative reconstruction methods, analytical-based reconstruction using projection weighting (PW)-based methods are commonly adopted to mitigate the truncation artifact via applying a weighting function on the doubly acquired redundant projection data. However, the disadvantage is the occurrence of an overall image intensity bias due to the missing data in the truncated projection region. Therefore, this work proposes evaluates the utilization of a modified complimentary rebinning (CR) approach as an alternative mitigation method while preserving the overall intensity of the reconstructed image. Using the recently developed MRI-based 4D CBCT digital phantom, five datasets are simulated based on the half-fan geometry to assess the performance of both PW and CR methods. The results indicated that CR effectively removes the truncation artifact with a significantly lower root mean squared error value at 0.18%, compared to PW at 0.51% (P=0.00). The evaluation of the scaling factor between the reconstructed and ground truth images indicated a reduction of half of the original overall intensity in PW with a mean value at 2.03, in contrast to CR at 1.01. Finally, using the tissue interface sharpness metric to quantify motion blur, no significant difference was found between PW and CR (P=0.23), even though CR displayed a higher value at 0.94 compared to PW at 0.90. This shows that the proposed CR method is able to produce an image with the same sharpness quality compared to the common PW in effectively mitigating truncation artifacts while preserving the overall intensity of the reconstructed image.