Study On CNN Material Decomposition Of Spectral CT And Helical Phase Contrast CT Reconstruction

Computed tomography(CT)is one of the most important breakthroughs in medical imaging technology in the 20th century.At the beginning of CT,the idea of energy spectrum CT had been proposed,which is to reconstruct the X-ray images in different energy bins independently.By setting different energy comparison thresholds,photon counting detectors can count and collect incident X-rays in multiple energy bins,have high energy resolution,and have good resistance to electronic noise,which is he important direction of research,and has made great progress in energy spectrum CT multi-material decomposition and phase contrast CT noise reduction.Spectral CT material decomposition includes projection domain decomposition,image domain decomposition and iterative reconstruction decomposition.Multi-energy spectral CT material decomposition often uses image domain decomposition method,which is also the focus of this paper.The image domain material decomposition method uses the reconstructed images of energy spectral CT of each energy bin to obtain the distribution images of various basis materials.Although the physical model of image domain material decomposition is relatively clear,in many cases,the reconstructed images of spectral CT inevitably have various noises,hardening,scattering,ringing artifacts,etc.,making the material decomposition equations highly ill-conditioning.It is difficult to obtain a good decomposition effect for traditional algorithm.In recent years,research on deep learning has made remarkable progress,and in many fields,it has achieved effects beyond traditional methods.Aiming at the advantage of the number of photon counting detectors,this paper introduces deep learning into material decomposition and designs a material decomposition method based on convolutional neural network.And in the simulation experiment and the real experiment,the error of the convolutional neural network material decomposition algorithm is one to two orders of magnitude smaller than the traditional image domain material decomposition algorithm.Phase contrast CT imaging and traditional transmission attenuation CT imaging have great complementarity.Experiments show that the phase change of soft tissue to X-ray is three orders of magnitude higher than the amplitude change.Therefore,soft tissue phase contrast imaging can achieve higher image contrast.The use of ordinary X-ray machine and grating technology is the mainstream of current phase contrast CT imaging research.After grating attenuation,the X-ray flux reaching the detector is greatly reduced,which is natural suitable for photon counting detectors.On the other hand,due to the low noise characteristics of photon counting detector,the quality of existing phase contrast CT images can be greatly improved.In this paper,for the cone-beam helicak phase contrast CT,the ”nearest ray” is used to extend the phase information extracted by the one-step exposure method into the three-dimensional cone beam helical CT imaging,which realizes the three-dimensional phase contrast CT image obtained by single helical CT scan.The simulation experiment was designed to verify the method.