An iterative approach for CT image reconstruction from low-dose data

Computed tomography represents a powerful imaging tool widely used in clinical practice. It is able to investigate the inner structures of the human body in a quantitative and non-destructive way. One of the most serious concern about this imaging methodology refers to the use of X-rays ionizing radiations, harmful for the patient. Even though the acquisition process can employ a reduced X-rays dose, conventional reconstruction algorithms are not able to deal with low-dose data (i.e., incomplete data). These approaches return images which quality is compromised, and consequently its clinical usability. This issue motivates the large interest of the research community in developing more effective approaches for image reconstruction in low-dose scenarios. Among various, deep learning solutions result the most widely adopted. However, their performance often strongly depend on the availability of a large and labeled training data set. This paper proposes a novel iterative approach that partially exploits some advantages of deep learning methodology but not requiring any training data set. As proved by preliminary results, the approach overcomes conventional reconstruction methods mitigating some computational issues and strongly reducing the artifacts that typically appear when a reduced number of views are employed during the acquisition.