Nonrigid Registration of Pre-Procedural MRI and Intra-Procedural CT in CT-Guided Cryoablation of Lung Tumors to Improve Lung Tumor Conspicuity

Abstract

To improve lung tumor conspicuity during CT-guided cryoablations, we used nonrigid image registrations to fuse pre-procedural MR images and intra-procedural CT images and determined which set of CT images taken at planning, targeting, and monitoring phases of the procedure provided the most accurate and fastest registrations. In 10 percutaneous CT-guided cryoablation procedures, MR images were registered with intra-procedural CT images using a nonrigid registration technique using an intensity-based approach with affine and B-Spline transformations. The time to complete the registration as well as the accuracy of the registration defined by Target Registration Error (TRE), Dice Similarity Coefficient (DSC), and Hausdorff Distance (HD) were measured to assess the performance of the registration. The least significant difference (LSD) method was used as a post-hoc analysis for comparing time and accuracy among planning, targeting, and monitoring phases. The mean TRE of the registration ranged from 6.26 (planning) to 10.31 (monitoring) mm. The mean DSC ranged from 83.86 (monitoring) to 89.22 (planning). The mean HD values ranged from 7.74 (targeting) to 12.20 (monitoring). Mean registration time ranged from 68.67 (monitoring) to 92.02 (planning) s. Using HD, registrations in either the planning or targeting phase were more accurate than in the monitoring phase. The registration was faster using monitoring images than using planning images. Nonrigid registration techniques can be used to fuse pre-procedural MR images with intra-procedural CT images with varying performance depending on the CT images taken at the different phases of the procedure. Therefore, caution should be taken in setting expectations on accuracies and speeds of registration depending on the phases of the CT-guided ablation procedures.