Dual-Channel Deep Convolutional Neural Network for Limited-Angle CT Image Restoration

Based on the excellent performance of computed tomography (CT) in visualizing the inside of objects, it has become one of the indispensable technologies in the fields of medical diagnosis and industrial inspection. However, due to uncontrollable factors such as radiation dose and detection environment, it is necessary to consider how to obtain projection information within a certain scanning rotation angle range in most cases. In these situations, the reconstructed images using traditional analytical algorithms and iterative algorithms will suffer from limited-angle artifacts. Recently, deep learning technology has gained great attention in the field of image restoration due to its powerful learning ability and its rapid development and application. In order to make up for the limitations of traditional methods in suppressing artifacts in limited-angle CT reconstruction, this paper designs a dual-channel deep convolutional network to remove artifacts in reconstructed images with preserving the details and edges. Simulated and real experiments demonstrate that the presented network can effectively restore images, compared with the classical network.

I. Introduction

X-ray computed tomography[1] is the most commonly used non-destructive testing technology in medicine and industry. It uses X-rays to perform tomography on the target and obtain the structural information of the target's internal tissues. In medicine, CT can detect diseases such as intracranial hemorrhage, fractures, cerebral infarction, and space-occupying lesions when diagnosing the human head. Chest CT can also detect organs such as the lungs and heart, and can determine whether the patient has pneumonia, lung cancer, pericardial effusion, etc. In industry, X-ray CT imaging can accurately reconstruct the internal structure of tubular objects and detect internal defects in materials. CT has a fast scanning speed and high resolution of scanned images, and has become an important imaging technology today.