上海科技大学知识管理系统

Photon-counting computed tomography (PCCT) is an industry-recognized next-generation CT technology. By utilizing a photon counting detector that provides information on object interaction that varies with different energy photons, physicians can differentiate materials in objects through material decomposition, enabling numerous medical applications such as detecting tumors or characterizing kidney stones. However, due to charge-sharing effects, application-specific integrated circuit (ASIC) pile-up effects, and Compton scattering in PCCT, there exists a mismatch between the real-world physical effects and the ideal assumptions used in the physics model, which can result in significant errors in material decomposition. To address this problem, this paper proposes a physics-guided deep-learning model for material decomposition. We construct a physics simulation model that can simulate the response of the PCCT system under different physics parameters and use it to build a training dataset. We then train a neural network on this dataset to learn the response of the CT detector and ASIC under different physics parameters. Next, we conduct a calibration experiment to adjust the parameters to reduce the difference between the neural network predictions and the actual data. During imaging, we calculate the thickness information of the detected substance by solving an optimization problem based on the physics parameters of the CT detector and the actual response of the CT detector. In summary, our proposed method addresses the problem of physical effects that deviate from the ideal physics model. Our method uses a small amount of experimental data and could be implemented in a real clinical setting.