Region-of-Interest reconstruction from truncated projection data under blind Object Support

Recently, the accurate Region-of-Interest (ROI) reconstruction from truncated projection data has been proved to be theoretically possible under few restrictions. One of these restrictions is the requirement of a priori knowledge of the object on a limited region inside the ROI. This region can be defined as a set of pixels having known intensity values inside or outside the object. In particular the knowledge of the background region outside the object is called the Object Support (OS). OS is defined as the region inside the scanner view where the object is certainly inside, therefore, the region outside the OS has zero intensity value. Using the OS constraint in image reconstruction when the projection data suffers from truncation has a large influence on image quality. The OS constraint is usually enforced by selecting a region slightly larger than the true object and set the pixel values outside this region to zero. However, the lack of the exact OS knowledge introduces DC-shift and low-frequency artifacts in the reconstructed image. In many clinical imaging scenarios, the prior knowledge of the exact OS is not available or requires extra efforts for an accurate estimation. In this paper, we propose an iterative reconstruction algorithm that automatically detects the OS during the reconstruction. The cost function for image reconstruction is modified by including a thresholding function in the form of the ℓ0 norm, the OS is estimated by the thresholding function and the OS converges toward the accurate one. Simulation results with different ROI and object configurations indicate a significant reduction in the artifacts by using the proposed algorithm.