Scatter and Beam Hardening Correction of Cone Beam CT with a Convolutional Neural Network

There are linear accelerators for the radiation therapy equipped with a cone beam CT (CBCT). The usage of this CBCT is for the confirmation of the position of an irradiation target and not for the planning of the radiation therapy. The reason is that measured CT data include lots of scattered photons due to the insufficient collimation of scattered photons. If we can use the CBCT for the treatment planning, we can make an adaptive therapy in which we make a modification of treatment plan according to the condition of a radiation target such as the size and position. The aim of this study is to remove the effect of scattered photons and beam hardening for the adaptive radiation therapy. To correct scattered photons and beam hardening effect, we used a convolutional neural network(CNN). The architecture of the CNN was the UNet++. To calculate cone-beam projection data of CT images, we conducted a Monte Carlo simulation with a cone-beam geometry and calculated projection data affected with scattered photons and beam hardening effect. These projection data were used for the input data of the CNN. To make training data without the effect of scattering and beam hardening, we calculated projection data with an average energy of x-rays. Simulation results showed that the image reconstructed with the projection data corrected by the CNN was close to the ideal image without scattering and beam hardening. These results showed the effectiveness of the proposed method.