Strategy to Reduce CT Dose by Using Graphic Processing Unit Based Iterative Reconstruction Algorithm- Preliminary Study

Abstract

Computed tomography (CT) has evolved into an indispensable imaging method in clinical routine for a long time. It gives us a high resolution, high contrast and noninvasive tomographic images in a brief scan time. But their high radiation dose is an important issue in CT system development.

Iterative algorithms can also give us better image quality than analytical types and reduce some artifact effectively. But they not only spend too much time, but also need very large scale memory size. In this study, we would like to reduce our projection data into <180-degree limited angle according to compressive sensing (CS) criteria and total variation (TV) minimization.

This is a computer simulation study combined with CPU and graphic processing units (GPUs). We developed a small scaled fully 3D system matrix, which included ray-tracing, voxel driven..., and circular redundancy. Two digital phantoms were used to simulate projection data sets and were reconstructed by post total-variation constrained algebraic reconstruction technique (ART-TV) algorithm and two statistical reconstruction techniques (SRTs): inner totalvariation constrained expectation-maximization (TV-EM) and post total-variation constrained expectation-maximization (EM-TV) in transmission types. To evaluate image quality, three indexes, root-mean-square error (RMSE), universal quality index (UQI) and coefficient of variation (CV) were used.

Our results shows that incorporation TV minimization in iterative reconstruction algorithm and followed by CS-criteria, we just need ultra-few projections to reconstruct the CT image to reach the same quality as the filtered back-projection (FBP) method with full-scan data sets. Besides, our study also demonstrates that SRT with <180-degree limited angle reconstruction can produce low-dose CT images.