Noise reduction in low-dose cone beam CT by incorporating prior volumetric image information

Purpose: Repeated use of cone beam CT (CBCT) in radiotherapy introduces extra imaging dose to patients. In this work, the authors propose a method to effectively reduce the imaging dose of on-treatment CBCT by incorporating previously acquired CBCT. Methods: The Karhunen-Loève (KL) transform was used to consider the correlation among the on-treatment low-dose CBCT and prior CBCTs in the projection domain. Following the KL transform, the selected CBCT projection data were decomposed into uncorrelated, ordered principal components. Then, a penalized weighted least-squares (PWLS) criterion was applied to restore each KL component using different penalty strengths, where the penalty parameter was inversely proportional to its corresponding KL eigenvalue. Following the inverse KL transform on the processed data, the FDK algorithm was used to reconstruct the on-treatment CBCT image. The proposed algorithm was evaluated using both phantom and patient data. Results: The proposed algorithm demonstrated the ability to suppress noise while preserving edge information effectively. This new strategy outperforms the PWLS algorithm without considering prior information based on the noise-resolution tradeoff measurement and analyze of the reconstructed small objects. Conclusions: Information extracted from previously acquired CBCT can be effectively utilized to suppress noise in on-treatment low-dose CBCT. The presented strategy can significantly lower the patient CBCT radiation dose.