2D/3D-MGR: A 2D/3D Medical Image Registration Framework Based on DRR

Abstract

Medical image registration is a crucial process in medical image analysis. However, traditional 2D and 3D medical image registration methods often struggle to accommodate the complex variations and conditions present in medical images. To address these challenges, this study introduces a novel 2D/3D image registration framework, termed 2D/3D-MGR, which leverages digitally reconstructed radiographs (DRR) to establish consistency between X-ray and computed tomography (CT) images. This research introduces the DCT-Net model for style transfer between DRR and X-ray images, and the CR-Net model for precise registration of intraoperative X-ray images with preoperative CT images. To validate the effectiveness of the proposed models, orthopedic X-ray and CT images were collected from Peking Union Medical College Hospital to create the XC dataset. Empirical results demonstrate that the DCT-Net model achieves a transformation quality level endorsed by medical experts, and the realistic style transfer is significant for subsequent image registration and surgical navigation. Additionally, the CR-Net model outperforms other state-of-the-art models in image registration accuracy, as evidenced by outstanding performance metrics on the XC spine dataset: a Pointwise Location Consistency (PLC) of 99.997% and a Root Mean Square Error (RMSE) of 0.1707. This novel image registration method significantly enhances the accuracy and efficiency of medical image processing, providing a valuable reference for future research in medical image registration.

Keywords

• Generative adversarial networks
• medical image registration
• image generation