SU‐FF‐I‐68: Evaluation of An Intensity‐Based Free‐Form Deformable Registration Algorithm

Purpose: To propose and evaluate an intensity‐based free‐form deformable registration algorithm for CTimage volumes. Method and Materials: We have developed an intensity‐based deformable registration algorithm with essentially limitless degrees of freedom depending on required accuracy and execution speed. The algorithm utilizes sufficient regularization to minimize the likelihood of folds or tears in the deformation fields. The first evaluation strategy utilized real CT volume pairs acquired weeks apart for three subjects who exhibited significant weight loss during IMRT treatment. For these subjects, we calculated the correlation coefficient after rigid and deformable registration. Additionally, a known deformation was applied to a real CT volume creating a synthetic target to determine the ability of the algorithm to recover the same “gold standard” deformation. Finally, we measured the consistency of the algorithm as the average distance of concatenated forward and reverse deformations from the unity transform on a real CT volume pair. Results: The correlation coefficients after rigid registration were 0.890, 0.921, and 0.859 for the three volume pairs. These improved to 0.979, 0.983, and 0.978, respectively, after deformable registration. These results were comparable to benchmark self‐correlations of the target CTs with a 1.4mm translation error applied (0.979, 0.980, and 0.978, respectively). In the second experiment, rigid registration averaged 10.3mm error (6.6mm SD) while deformable registration averaged 1.1mm error (1.9mm SD) from the “gold standard”. Nearly three quarters (73.9%) of voxels had less than 1mm error and the 95% confidence interval was 4.8mm compared to 0.6% and 23.2mm respectively for rigid registration. Finally, concatenating the forward and reverse deformations resulted in an average distance of 3.1mm (3.1mm SD) from unity. Conclusions: The accuracy and speed of the algorithm suggests its utility for a variety of purposes including integration into an adaptive radiation therapy protocol. Conflict of Interest: Contributing author employed by MIMvista Corp.