Reconstruction Of Helical Cone-beam CT System And Deep Learning Based Beam Hardening Artifact Reduction Method

Computed Tomography(CT)is a non-destructive testing technology to observe slices ofobjects and has been widely used in medical imaging and industrial testing field.Compared with traditional testing method such as X-ray photographs and digital radiography,slices reconstructed by CT have the advantages of high spatial resolution and contrast.CT images have no overlapping regions of testing object and it is easy to see the internal defects of the object.So we can get tomography of good quality without open up the object.Helical cone-beam CT has the advantages of high scanning speed and high accuracy.Helical cone-beam CT is now widely used in medical diagnostic field but not in industrial testing field.Helical cone-beam CT can scan the object without stop and this makes helical cone-beam CT more effective for the scanning of long workpiece.So the implementation of helical cone-beam CT to industrial testing field has a practical significance.The main work of the implementation of helical cone-beam CT in this paper mainly contains three parts: the implementation of classical Katsevich reconstruction algorithm for helical cone-beam CT and its parallel accelerated implementation,design for related software,design and implementation for three-dimensional visualization software.The actual energy spectrum of X-ray is continuous but most of the reconstruction algorithm are based on the assumption that the X-ray is mono-energetic,which results in beam hardening artifacts in the reconstructed tomographic image.Beam hardening artifacts greatly reduce the quality of the image,not only affect the diagnostic results,but also have an impact on the subsequent automated size measurement and image segmentation et al.Hence it is of great significance on how to eliminate the beam hardening artifacts.Most of the existing beam hardening artifact elimination methods usually need to establish a model to correct the polychromatic projection to monoenergetic projection or correct the slices with artifacts to slices without artifacts.Although these existing methods are effective to some extent,the correction process is mostly complicated and incomplete.In this paper,a beam hardening artifacts correction method based on deep learning is proposed for homogeneous objects.The beam hardening artifacts in the reconstructed image are displayed as cupping artifacts and streak artifacts,and the artifacts have a certain regularity.Deep learning can fit the complex relationship between input and output,and it also can learn the features of the image and combine them together.Hence this paper uses deep learning to study the relationship between images with beam hardening artifacts andartifact images to suppress beam hardening artifacts.The input of the network are slices reconstructed by FBP algorithm with polychromatic projection,and the output are the ideal reconstructed slices at a certain energy.The simulation results show that the proposed method is effective and has a certain applicability to the different energy spectrum.