Identifying Sets of Favorable Projections for Few-View Low-Dose Cone-Beam CT Scanning

— Cone-beam computed tomography is beginning to emerge as a widely used technique for medical imaging. However, here has been growing concerns with regards to the X-ray dose delivered to the patient. Low dose CT has thus been gaining substantial interest. It can be achieved by lowering the X-ray dose per projection and/or reducing the number of projections acquired. In this work we focus on the latter and provide some new insight on how to identify favorable (or salient) views that maximize the information used within an iterative reconstruction framework. Based on prior knowledge on the object to be scanned, we propose an optimization framework that can automatically identify a minimal set of projections that can capture the salient object features. Our results indicate that these generalized views result in better image quality than evenly distributed projections. I. INTRODUCTION Cone-beam CT has emerged as a major X-ray imaging modality both in terms of image quality and scan time. A popular cone-beam CT reconstruction method is the FDK [1] algorithm, which provides high resolution results but requires several hundreds of patient X-ray projections. With the growing concern about the potential risk of X-ray radiation exposure to the human body, dose reduction in cone-beam scanning (and other modalities) has become a significant research topic. Dose reduction usually involves lowering the X-ray energy per projection and/or reducing the total number of projections. Both methods typically suffer from low signal-to-noise ratio (SNR) in the reconstructions. Iterative reconstruction schemes, matched with suitable regularization methods were shown to cope well with these few-view or high-noise scenarios [2][3][4]. The work presented here focuses on one specific low-dose CT measure: reducing the number of projections. It capitalizes on the fact that in standard radiography physicians and X-ray technologists typically have a good idea, often based on standards, at what patient orientation the radiograph should be taken to reveal the desired insight. We denote these views as salient views. We propose to formalize and generalize the concept of salient views for CT reconstruction, and use iterative CT reconstruction to cope with the potentially irregular and sparse view distribution. To identify the salient views we analyze prior reconstructions, locate the salient features, and determine the projection(s) at which these features differentiate best. Once the salient views are obtained, we use a set-covering framework to accelerate the search for the optimal scanning configuration and trajectory that covers all of these views.