A Multi-path Global and Local Cycle GAN for Enhancing CT Brain Image Synthesis from MR Brain Images

Hidemasa Hayashi, Yutaro Iwamoto, Masahiro Nonaka, Yen-Wei Chen

Abstract: CT and MRI are essential for the diagnosis and effective treatment of various diseases. However, acquiring these images separately is a time-consuming and costly process for both patients and physicians. Additionally, CT scans contribute to increased radiation exposure for patients. Therefore, generating CT scans from radiation-free MR images is highly desirable. Synthetic brain CT scans are valuable for obtaining crucial information for neurosurgical procedures and treatments, although they are challenging to derive directly from MR images. In this study, we propose a multi-path global and local cycle GAN (GL-CycleGAN), which is an improved CycleGAN based approach. By incorporating Global and Local modules in the generator, we extract both overall structural features of the cranium and detailed characteristics of local regions, enabling smoother representation of cranial curvature and finer details reproduction, thus achieving higher precision in synthetic results.