3D multi-modality medical image registration with synthetic image augmentation using CycleGAN

Abstract

This report proposes a 3D multi-modality medical image registration network with CycleGAN-based synthetic image augmentation. The method is designed for intra-subject brain CT-MRI registration. A broad overview of our method is to first generate a synthetic CT image from the MRI using the CycleGAN and then align it with the MRI using the registration network to learn a deformation field which is then used to register the MRI with the CT. Furthermore, we use image-to-image similarity metrics between the synthetic CT and the CT along with an additional auxiliary loss between the warped MRI and the CT. Finally, we perform thorough experiments on our method and prove that it outperforms the state-of-the-art methods and tools.