| Deformable registration methods for Cone Beam CT (CBCT) guided surgical procedures are developed and experimentally validated. Use of deformable registration for guidance of therapeutic procedures places tradeoffs on accuracy and computational expense; in addition, existing algorithms may not be robust in the face of challenges faced during CBCT-guided procedures, such as image intensity inaccuracy or the surgical removal of tissue between image acquisitions. The popular Demons registration algorithm is investigated and extended with a novel convergence criterion, an iterative tissue specific intensity match procedure, and a method denoted extra-dimensional Demons (XDD) is developed to perform registration in the presence of tissue excision or the introduction of new material.;The proposed extensions to the algorithm are tested in phantom, cadaver, and patient images and found to preserve the speed and simplicity of the Demons algorithm while maintaining or improving accuracy and robustness. Multi-resolution demons registration combined with the appropriate convergence criterion achieved target registration error (TRE) = (0.8 +/- 0.3) mm and normalized cross correlation (NCC) = 0.99 within one minute for cadaver data obtained using a prototype CBCT capable C-arm, compared to TRE = (2.6 +/- 1.0) mm and NCC = 0.93 with rigid registration. The iterative intensity matching approach proved robust against examined levels of intensity variation as measured in cadaver data and six patient cases obtained from CBCT-guided surgery and CBCT-guided radiation therapy. In the presence of content mismatch between images, phantom experiments showed the normalized mutual information (NMI) in regions local to the excision to improve from 1.10 for the conventional Demons approach to 1.16 for XDD, and qualitative examination revealed the conventional Demons approach imparted unrealistic distortions in areas around tissue excision, whereas XDD provided accurate "ejection" of voxels within the excision site and maintained the registration accuracy throughout the rest of the image.;Fast, accurate, deformable image registration is an important aspect of image-guided interventions. The work presented in this dissertation aims to make deformable registration applicable for use in CBCT-guided surgical procedures by presenting a fast, accurate method of deformable registration that is robust in the face of challenges presented during CBCT-guided interventions. |
