X-ray Multi-energy CT Based On Blind Source Separation Of Multi-voltage Image Stack

Compared to conventional computed tomography(CT), which is based on the assumption of monochromatic X-ray, multi-energy CT includes more information of the spectrum to an mapping between energy and linear attenuation coefficients. Different components of imaged objects can be distinguished effectively. It meets requirements of the functional CT which is applied to component microstructure quantitative analysis of new materials, mineral processing and modern medicine. So far, there are still some limitations of temporal and spectral resolution for dual-energy CT and multi-energy CT based on photon-counting detectors or spectrum filtering separation. Synchrotron radiation emits monochromatic X-rays, but as a national large scale science facilities, it is shared broadly with limited machine-hour for each sample tests. Thus, test efficiency and scale is influenced. In this paper, the decomposition of X-ray image stack with polychromatic spectrum is researched to expect the acquisition of images comparable to a narrow-energy-width case by no change of hardware in the usually CT imaging system. Then the effective distinguish of different components can be implement in the usually CT imaging system, which supports quantitative characterization of microstructure.A decomposition method of multi-voltage polychromatic X-ray images is proposed with the analogy of the problem and blind source separation based on the analysis of polychromatic X-ray imaging and monochromatic CT. The decomposition model is built by to minimizing the sum of squared decomposition error. The solving algorithm is derived by imitating the non-negative matrix factorization. For its local convergences, the genetic algorithm, which has globle convergence, is embed in the solving algorithm. The reconstructed images from the decomposed projections show some characteristics with narrow-energy-width images.For the influence of scattering in practical CT imaging, the local variance sum is used to describe the low-frequency characteristics of scattering. Then an improved model is proposed with the optimization goal to minimize the local variance sum of decomposition residual. The decomposed result is improvement compared to the first model.For the disorder of decomposed projection and uncertainty of spectrum partition, a method of narrow-energy-width energy calibration is proposed by replacing the linear attenuation coefficients with its decomposition according to photoeffect and compton scattering. Combining physical prior information, some CT images reconstructed from decomposed projections are fused with the data-constrained model(DCM) to quantify the characterization of components.In this paper, both theoretical analyses and compared experiments are studied. A cylinder composed of aluminium and silicon is used in simulation and practical experiments to verify the feasibility of polychromatic X-ray images decomposition, the first model, the improved model, the narrow-energy-width energy calibration and CT images fusion. The experiment results show that the proposed method of multi-energy CT based on blind source separation can achieve quantitative characterization of components by innovating scanning mode and data processing method with no change of hardware in the usually CT imaging system. It has important meaning and application value to improve test efficiency, reduce research cost and promote the development of multi-energy X-ray CT imaging and quantitative characterization.