Spectral computed tomography (CT) with photon counting detectors (PCDs) can collect photons by setting different energy bins. It is well acknowledged that PCD-based spectral CT has great potential for lowering radiation dose and improve material discrimination. One critical processing in spectral CT is energy spectrum modelling or spectral information decomposition. In this work, we proposed a dual-domain deep learning (DDDL) method to calibrate a spectral CT system by a neural network. Without explicit energy spectrum and detector response model, we train a neural network to implicitly define the non-linear relationship in spectral CT. Virtual monochromatic attenuation maps are synthesized directly from polychromatic projections. Simulation and real experimental results verified the feasibilities and accuracies of the proposed method.
Spectral computed tomography (SCT) has advantages in multienergy material decomposition for material discrimination and quantitative image reconstruction. However, due to the nonideal physical effects of photon counting detectors, including charge sharing, pulse pileup and K-escape, it is difficult to obtain precise system models in practical SCT systems. Serious spectral distortion is unavoidable, which introduces error into the decomposition model and affects material decomposition accuracy. Recently, neural networks demonstrated great potential in image segmentation, object detection, natural language processing, etc. By adjusting the interconnection relationship among internal nodes, it provides a way to mine information from data. Considering the difficulty in modeling SCT system spectra and the superiority of data-driven characteristics of neural networks, we proposed a spectral information extraction method for virtual monochromatic attenuation maps using a simple fully connected neural network without knowing spectral information. In our method, virtual monochromatic linear attenuation coefficients can be obtained directly through our neural network, which could contribute to further material recognition. Our method also provides outstanding performance on denoising and artifacts suppression. It can be furnished for SCT systems with different settings of energy bins or thresholds. Various substances available can be used for training. The trained neural network has a good generalization ability according to our results. The testing mean square errors are about 1 × 10 − 05 cm − 2.
Spectral Computed Tomography (CT) has an advantage of providing energy spectrum information, which is valued on multi-energy material decomposition for material discrimination and accurate image reconstruction. However, due to the non-ideal physical effects of photon counting detectors (PCDs), such as charge sharing, pulse pileup and K-escape, serious spectral distortion is unavoidable in practical systems. The degraded spectrum will induce error into the decomposition model and affect the accuracy of material decomposition. Recently, artificial neural network has demonstrated great potential in the tasks of image segmentation, object detection, natural language processing, and etc. By adjusting the interconnection relationship among a large number of internal nodes, a neural network provides us a way to mine information from huge data depending on the complexity of the network system. Considering the difficulty of modeling the spectral CT system spectrum including the response function of a PCD and the superiority of data-driven characteristics of a neural network, we proposed a novel multi-energy material decomposition method using a neural network without the knowledge of spectral information. On one hand, specific linear attenuation coefficients can be obtained directly through our method. It would help further material recognition and spectral CT reconstruction. On the other hand, the network outputs show outstanding performance on image denoising and artifacts suppression. Our method can fit for different selections of training materials and different settings of imaging systems such as different number of energy bins and energy bin thresholds. According to our test results, the trained neural network has a good generalization ability.