CT Reconstruction from Few-Views by Anisotropic Total Variation Minimization'

Dose reduction is an important factor in medical imaging, One of the key strategies to reduce the dose in CT reconstruction is under-sampling, T he recently proposed Compressed Sensing (CS) paradigm has shown that nearly exact CT reconstructions can be obtained from highly under-sampled projection sets using e. g. regularized iterative reconstruction algorithms. For the regularization, the standard Total Variation (TV) minimization method is commonly applied. T he TV has a sparse setting and can therefore be utilized in the CS framework. It can smooth undesired discontinuities related to noise and preserve edges of the field-of-interest objects. It has developed a de facto standard method in CS based CT reconstructions. Nevertheless, the utilization of the Total Variation method has several drawbacks. One of these is over-smoothing of some fine objects, structures and textures and at the same time great spatial resolution loss. T he purpose of this work is to develop a new A nisotropic Total Variation (ATV) method that can overcome these disadvantages and produce strong results in the sparse environment. In this work it will be shown that the usage of ATV regularization leads to edges, objects and fine structures which are sharper and clearer in contrast to the standard TV method. In addition to that the spatial resolution for the ATV method is much higher than for the TV method. T he performance of both methods will be evaluated visually by means of digital simulated data, as well as real data scans obtained from an Elekta cone-beam synergy system.