Synthetic CT generation from CBCT using double-chain-CycleGAN.

•A new network named DCC-GAN is proposed to improve the quality of CBCT-based synthetic CT generation. DCC-GAN adopts a multi-resolution backbone network structure and a new generic feature extraction module, which is lightweight under the premise of performance improvement, and the training and inference time are shorter.•A learning rate adjustment strategy based on the loss value of the generator and discriminator is proposed, which improves the training stability of the GAN network, resulting in performance improvements.•For some disadvantages brought by the second point, such as the influence of noise on the model is amplified, an improved loss function is proposed, which effectively improves the noise reduction ability and training stability of DCC-GAN.