A CNN based volumetric imaging method with single X-ray projection

The accurate volumetric imaging using breathing motion model and single X-ray projection has a significant meaning to the clinical treatment of moving tumors. However, it is of great difficulty to realize the real-time reconstruction of volumetric image with high accuracy. In this work, a volumetric imaging method from one X-ray projection utilizing convolutional neural network (CNN) is proposed. With the aid of principal component analysis (PCA)-based motion model, a CNN regression model with weighted object function is trained to estimate the volumetric image accurately. Due to the high parallelization of CNN, the computing efficiency of the proposed method is able to meet the real-time requirement of practical treatment (less than 0.05 seconds). A synthetic test using 4D Extended Cardiac-Torso (XCAT) Phantom is carried out which testify the effectiveness of our method.