Research On Spectral CT System Calibration And Multiple Information Analysis Optimization Method

The technology of Computed Tomography(CT)has palyed a vital role in medical imageing,security check and industrial nondestructive testing.However,the facts of the low denstity resolution and beam hardening artifacts in tradional CT and of the overlapping of energy spectrum in dual-energy CT have limited in the application in CT.Photon counting detector(PCD)with high count rate and high resolution,collecting photons by setting different energies,provides CT with spectral information.Spectral CT,in hence,is more and more widely applied in the filed of low-dose imaging,quantitative imaging and multienergy material decomposition,etc.However,due to non-ideal physical effects of PCD,spectrum may be distorted unavoidebly,which has an impact on performance and applications of spectral CT.In this paper,by focusing on these problem,we carried out researches on spectral CT system calibration,including the energy calibration of PCD and the spectrum calibration of the system with multiple information analysis optimization methods.Meanwhile,the spectral information is accurately extracted and quantitative CT imaging is realized.Spectral CT hopefully could be put into application on a large scale.The energy calibration of PCD is important and fundamental before all processing and applications of spectral CT.Firstly,we porposed a calibration method for PCD energy calibration using only one monochromatic source in Timee-over-Threshold(TOT)counting mode.By establishing an implicit parameter set of energy deposited in each detector element within a local region under each photon detection event,we define an objective function according to the statistical distribution of accumulated energy in each event.The relationship of TOT counts and energies under each digital to analog converter(DAC)thresholds of PCD is modeled by a non-linear model.Through DAC threshold scaning,the relationship of DAC thresholds and enegies could be determined by optimize the objective function.This method requires no detector response model and naturally take advantage of the energy spreading among neighboring detector elements to acquire information of detector response from various energies.The practical experiments are simple and easy to operate.The results show the feasibility,applicability and accuracy of our method.To solve the spectrum calibration problem for a spectral CT system,we proposed three methods: an empirical model by polynomial fitting and two models base on neural network.By building models between polychromatic CT projection raw data or polychromatic CT reconstructions and virtual monochromatic attenuation maps using polynomial functions and implicit functions trained by neural network,the spectral information extraction and CT reconstruction were completed.Parametes in the polynomial model and neural network models were determined by minimizing an ensemble cost function in image domain,which provide us a great convenience in selecting calibration materials and constructin calibration phantoms Through numerical simulations and practical experiments,the feasibilities of the methods were verified.We quantitatively analyzed the mean relative error(MRE)and mean square error(MSE)of each material in virtual monochromatic attenuation maps obtained from three models.Finally,we compared the accuracy,applicability and performace on noise denoising and artifact suppression of all mehods.