ROI-Based Intraoperative MR-CT Registration for Image-Guided Multimode Tumor Ablation Therapy in Hepatic Malignant Tumors

Image-guided ablation therapy has been widely used as a minimally invasive treatment for hepatic malignant tumors. Image registration is very useful during such operations, especially for multimode tumor ablation therapy. This research proposes a novel idea for performing MR-CT registration in multimode tumor ablation therapy based on liver segmentation. First, the liver is segmented from the preoperative 3D MR image and the intraoperative 3D CT image using a deep learning method based on a modified UNet++ architecture. Then, the preoperative MR image and the intraoperative CT image are coregistered using rigid and nonrigid registration methods with the segmented liver as the region of interest. The segmented binary images, rather than gray-level images, are aligned in the rigid registration step, which proves to be faster and more accurate than the registration method based on gray information. For the nonrigid registration, a multilevel free-form deformation method is applied to correct tiny misalignments. Finally, our method was validated using clinical data from 15 patients. The proposed method achieved an average Dice coefficient and target registration error of 93.36±1.21% and 4.42±2.35 mm, respectively, and it can help interventional radiologists adjust the probe position in clinical work.