Prospective Regularization Analysis and Design for Prior-Image-Based Reconstruction of X-ray CT

Hao Zhang,Hao Dang,Grace J. Gang,J. Webster Stayman

2017-01-01

Abstract:Prior-image-based reconstruction (PIBR) methods, which incorporate a high-quality patient-specific prior image into the reconstruction of subsequent low-dose CT acquisitions, have demonstrated great potential to dramatically reduce data fidelity requirements while maintaining or improving image quality. However, one challenge with the PIBR methods is in the selection of the prior image regularization parameter which controls the balance between information from current measurements and information from the prior image. Too little prior information yields few improvements for PIBR, and too much prior information can lead to PIBR results too similar to the prior image obscuring or misrepresenting features in the reconstruction. While exhaustive parameter searches can be used to establish prior image regularization strength, this process can be time consuming (involving a series of iterative reconstructions) and particular settings may not generalize for different acquisition protocols, anatomical sites, patient sizes, etc. Moreover, optimal regularization strategies can be dependent on the location within the object further complicating selection. In this work, we propose a novel approach for prospective analysis of PIBR. The methodology can be used to determine prior image regularization strength to admit specific anatomical changes without the need to perform iterative reconstructions in advance. The same basic methodology can also be used to prescribe uniform (shift-invariant) admission of change throughout the entire imaging field of view. The proposed predictive analytical approach was investigated in two phantom studies, and compared with the results from exhaustive search based on numerous iterative reconstructions. The experimental results show that the proposed analytical approach has high accuracy in predicting the admission of specific anatomical features, allowing for prospective determination of the prior image regularization parameter.