Research On X-ray Energy-resolving Technology For Spectral CT Image Reconstruction

Computed Tomography(CT)is an important technology in the field of clinical diagnosis and has a broad development prospect.Spectrum CT provides many quantitative calculation methods and functional imaging approaches for the identification of substances,solving the problem of “different substances with the same CT value” in traditional CT,but also has higher requirements for the accuracy of energy-resolving and reconstruction process.Aiming at the accuracy of energy-resolving process and spectral imaging,this thesis establishes an energy deposition model of multi-layered pixel structure based on Monte Carlo method,and uses energy deposition coefficient instead of traditional energy attenuation coefficient realizing the multi-energy resolving in the clinical diagnosed X-ray range of 15keV-140 keV.After analyzing the noise of resolving process,a reconfigurable energy-resolving method is proposed.By setting different spectrum thresholds and using the accumulation of charge distribution within different thicknesses,the systematic and random errors in energy resolving method is reduced,improving the flexibility and accuracy of spectral imaging.Aiming at the problem of beam hardening artifacts in the reconstruction process,a post-processing correction method based on Convolutional Neural Network(CNN)was proposed.The method uses CNN to extract information of low-energy and high-energy grayscale images simultaneously to enhance the ability of identifying and correcting artifacts and improve the image quality of dual-energy post-reconstruction method.Under different threshold settings,two energy bins have been dynamically resolved with the reconfigurable energy-resolving method.By the resolved spectrum,dual-energy reconstruction process has also been completed through a self-built phantom.Based on a self-built phantom library with five elements,the experiment of beam hardening artifact correction is completed with two spectrum threshold settings.The experiment results show that the average value of the relative error in the reconfigurable energy resolving process is 5.65% which is far lower than that with traditional energy attenuation coefficient.The mean PSNR value of the effective atomic number and electron density image is 32.65 dB and 47.79 dB,respectively.In the two sets of artifact correction experiments,the PSNR value of the corrected effective atomic number images was improved by 3dB and 6dB,and the electron density image was improved by 0.7dB and 1.5dB.All the "cup artifact" and "Strip artifacts" in images were significantly suppressed.In summary,the reconfigurable energy-resolving method can improve the accuracy of resolving results and the flexibility of CT imaging system,reducing the interference of scattering noise between layers and random noise for reconstruction images;the proposed beam hardening artifact correction method can further improve the quality of reconstruction images for clinical diagnosis.