Interweaving Network: A Novel Monochromatic Image Synthesis Method for a Photon-Counting Detector CT System

Abstract

With the growing technology of photon-counting detectors (PCD), spectral CT is an important topic for its potential in material differentiation. However, direct reconstruction of the detected spectrum without any compensation will lead to inaccurate results due to some non-ideal factors such as cross talk and pulse pile-up in the detectors. Conventional methods try to model these factors using calibrations and make compensations accordingly, but the results depend on the model calibration accuracy. In this paper, we proposed an Interweaving Network (WeaveNet), a novel deep learning-based monochromatic image synthesis method working in sinogram domain. Unlike previous deep learning-based methods, the WeaveNet architecture was designed based on the factor of spectrum distortion and it can solve this problem better in an intuitive way. The method was tested on a cone-beam CT (CBCT) system equipped with a PCD. After FDK reconstruction of the synthesized monochromatic projection, we evaluated the accuracy of linear attenuation coefficient, decomposition coefficient and separation angle of different materials to examine the performance of our method. This method gives more accurate results with less noise than previous methods, which demonstrates the advantages of this monochromatic image synthesis method.