CBCT-Based Synthetic MRI Generation for CBCT-Guided Adaptive Radiotherapy

摘要

Cone-beam computed tomography (CBCT) has been widely used in image-guided radiation therapy for patient setup to improve treatment performance. However, the low soft tissue contrast on CBCT may limit its utility when soft tissue alignment is of interest. Moreover, the potential application of CBCT in adaptive radiation therapy also requires superior soft tissue contrast for online target and organ-at-risk delineation and localization. The purpose of this study is to develop a deep learning-based approach to generate synthetic MRI (sMRI) from CBCT to provide a high soft tissue contrast on CBCT anatomy. The proposed method integrates a dense block and self-attention concept into a cycle-consistent adversarial network (cycleGAN) framework, called attention-cycleGAN, to learn a mapping between CBCT images and paired MRI. Compared with a GAN, a cycleGAN includes an inverse transformation from CBCT to MRI, which constrains the model by forcing a one-to-one mapping. A fully convolution neural network (FCN) with U-Net architecture is used in the generator to enable end-to-end CBCT-to-MRI transformations. Dense blocks and self-attention strategy are used to learn the information to well represent the CBCT image and to map to the specific MRI structure. The experimental results demonstrated that the proposed method could accurately generate sMRI with a similar soft-tissue contract as real MRI.