Deep-Learning-Based Denoising Of X-Ray Multi-Contrast Images

In recent decades,with the breakthrough in integrated circuit technology,the computing power of computers has been greatly enhanced,and deep learning technology has also developed rapidly.In the field of imaging,deep learning technology has achieved excellent image noise reduction in many interdisciplinary subjects.In the field of X-ray imaging,grating-based X-ray phase contrast imaging(GBPCI)is generally regarded as one of the most promising imaging techniques,it is compatible with conventional X-ray sources and realizes multi-contrast imaging of soft tissues.It is widely used in clinical medicine,materials science and Industrial non-destructive testing.The quality of multi-contrast images is often disturbed by shot noise in GBPCI,especially under the condition of low signal-to-noise ratio.In addition,while in principle the phase map could be obtained through one-dimensional integration of the differential phase image,this procedure is known to produce strong streak artefacts along the integration direction,due to propagation of the noise in the refraction image.In view of the above problems,this dissertation aims to improve the quality of multi-contrast images and optimize the potential radiation dose,and mainly carry out the following work:1.Conducted research on deep denoising of X-ray multi-contrast images.In order to solve the problem that multi-contrast images are disturbed by Poisson noise,this dissertation presents the convolutional neural network Dn CNN-P and evaluates its noise reduction performance.Two methods of " retrieval-denoising " and " denoisingretrieval " were designed,and the performance of the two denoising methods under different experimental conditions was evaluated and compared through simulation experiments and biological sample experiments.Experimental results show that with the algorithm Dn CNN-P used,the D-R mode always exhibits a better noise reduction under diverse experimental conditions,even in the case of a low photon count and/or a low visibility.For example,with a detected photon count of 1800 and a visibility of 0.3,compared to the differential phase images without denoising,the standard deviation is reduced by 89.1% in the D-R modes.This novel algorithm can be a promising approach to improve the quality of X-ray differential phase and dark-field images,and therefore dose efficiency in future biomedical applications.2.Based on the quasi-homogeneous material approximation,the retrieval of X-ray phase images and the study of noise properties are carried out.Under the approximation of quasi-homogeneous material,a direct retrieval method of X-ray phase images was developed by using the constant approximation law of the ratio of the real part to the imaginary part of the complex refractive index,and a verification experiment was carried out at the Shanghai Synchrotron Radiation Facility.Experimental results confirm the accuracy of direct retrieval of phase images.Compared with the retrieved phase image,the noise of the directly retrieved phase image is significantly suppressed,and the streak artifact is significantly weakened.3.Based on the phase-attenuation duality of soft tissues in high-energy X-ray imaging,the retrieval and noise properties of X-ray phase images were carried out.Using the complementary relationship between the phase and attenuation of soft tissue samples under high-energy X-ray illumination,a method for direct retrieval of phase images using a single projection image is investigated.On the basis of theoretical derivation and extraction formula,numerical simulation experiment research was carried out.The results confirmed the feasibility and accuracy of direct retrieval of phase images.Further noise analysis results show that the sensitivity of direct retrieval of phase images is consistently higher than that of traditional integrated phase images.