X-ray Spectral CT Image Quality Optimization And Material Discrimination Study

X-ray Computed Tomography(XCT)can obtain a tomographic image that reflects the internal structural information of the measured object.As a non-invasive and noncontact detection,XCT has achieved outstanding results in the fields of modern medical clinical diagnosis and industrial instrument quality inspection.Nowadays,most of the X-ray detectors used in CT systems are still energyintegrated ray detectors,which ultimately collect the average attenuation information over the entire ray energy spectrum.In contrast,the spectral CT system based on X-ray photon counting detector,which the detector is able to classify the photons in different energy ranges,can obtain the X-ray attenuation characteristics of the measured object in different energy ranges.However,due to the inconsistency of the photon response of the X-ray photon counting detector and the limited number of detecting photons in some narrow energy spectral ranges,there are still certain noise and artifacts in the images obtained from the spectral CT system,and the image quality seriously affects the discrimination and decomposition study of material components in spectral CT images.In recent years,with the booming development of artificial intelligence technology,research in the related field of spectral CT has been flourishing with the help of deep learning theory.The research work of this thesis relied on the National Key Research and Development Program(Project No.2016YFC0104609),the Fundamental Research Funds for the Central Universities(Project No.2019CDYGYB019)and the Natural Science Foundation of Chongqing(Project No.Ncstc2020jcyj-msxm X0553),and conducted research on spectral CT image quality optimization and material discrimination using deep learning theory.The main research content of this thesis is as follows:(1)On the basis of introducing the imaging theory of conventional XCT,the working principle of photon counting detector is systematically described,and summarizing the features and advantages of spectral CT imaging technology,so as to lay the corresponding theoretical foundation for the later research of spectral CT image quality improvement and spectral CT image material discrimination.(2)For the problem of noise and artifacts in spectral CT images,this thesis designs a neural network model for improving the quality of spectral CT images with the help of deep learning techniques.The mouse specimen was scanned to construct a complete spectral CT images dataset by spectral CT system,and comparative experiments were conducted to verify the feasibility of the network model.The experimental results show that the proposed network is effective for spectral CT image denoising and artifact removal.(3)A neural network model for spectral CT image-domain material decomposition is proposed in combination with generative adversarial networks for the problems of low decomposition accuracy and low decomposition efficiency of traditional decomposition methods in the study of spectral CT image-domain material discrimination.The comparison experiments were designed to verify the feasibility of the method,and the test results of each network were analyzed and the evaluation indexes of each network were compared.The research results show that the method can effectively improve the efficiency and accuracy of spectral CT image material decomposition.