2124 - Deformable Image Registration for Assessing Time-Course Effect of Proton Beam in the Heart in the Presence of Large Anatomical Variations
Presenter(s)
H. Kim1, F. J. Bijari1,2, J. Kang1, Y. Chemli2, N. Depauw1, A. Garonna3, S. H. Moon1, Y. L. E. E. Chen1, R. A. Levine1, H. I. Lehmann1,4, M. Normandin2, G. El Fakhri2, T. R. Bortfeld1, and K. W. Jee1; 1Massachusetts General Hospital, Boston, MA, 2Yale University, New Haven, CT, 3EBAMed SA, Geneva, Switzerland, 4University of Utah, Salt Lake City, UT
Purpose/Objective(s): Radiation-based ablation is a promising alternative to catheter-based methods for treating ventricular tachycardia. Optimizing treatment and evaluating the response to therapy is crucial for the success of these interventions. An animal study was conducted to assess changes in the myocardial tissue membrane potential at various time points using PET-CT scans following proton beam irradiation. The rapid growth of swine and structural changes around target regions after irradiation complicated the effective propagation of contours to follow-up imaging time points using a conventional deformable image registration (DIR). To enhance the robustness of contour propagation, a novel DIR method has been developed.
Materials/Methods: Six swine underwent proton beam irradiation and longitudinal PET-CT scans using [18F](4-Fluorophenyl)triphenylphosphonium ([18F]FTPP+). The contours for the left ventricle (LV) and the target area were delineated on the baseline 4D contrast-enhanced CT images at 80% of the cardiac cycle during end-expiration. DIR and contour propagation utilizing the calculated deformable vector field (DVF) was performed on the 4D contrast-enhanced CT images acquired at 2, 8, and 14 weeks post-irradiation. The DIR method involves four steps: first, approximating overall growth based on an affine transformation; second, narrowing down the region of interest; third, calculating cardiac motion vector fields at given time points; and fourth, calculating the DVF between images acquired at adjacent time points. For DIR, a conventional intensity-based image similarity metric was modified to include an image-derivative similarity term. To have a topology-preserving DVF, the Jacobian determinant of the DVF was regularized to ensure positive values, and the boundary values of the DVF were minimized to zero. Additionally, DVFs were further constrained with inverse consistency and cardiac-motion consistency regularization terms.
Results: The proposed motion-informed image-derivative-based DIR (MIDIR) method successfully aligned structures of interest for six subjects. Implausible deformations due to anatomical changes near hearts were significantly reduced compared to the conventional intensity-based DIR method.
Conclusion: We have developed a novel DIR framework to propagate cardiac contours in the presence of substantial changes in subject anatomy. The developed method successfully describes the physiological growth and deformation of cardiac structures at different time points, even when significant treatment-related changes are made in the structures of/near the heart.