Purpose: Cone-beam CT (CBCT) suffers from intensity inhomogeneities, which makes deformable image registration less reliable. We evaluate the use of intensity correction pre-processing to improve registration for adaptive radiotherapy in prostate cancer.
Methods: Intensities in the CBCT are corrected in a patch-based approach using a rigidly aligned planning CT as reference. Rigid registration is performed using mutual information (MI), and a linear gray-level transformation of intensities are calculated from their mean and standard deviation within small patches in both images. Blending is applied between patch centers to obtain a smooth correction field. Following intensity correction, a mean squared error (MSE) similarity metric is used to optimize a multi-resolution B-spline registration. Ten prostate cancer cases are evaluated, by comparing propagated contours of prostate, rectum and bladder against reference contours. In addition, distances of implanted gold seeds in CBCT and warped CT are obtained for seven patients and two fractions (3-5 seeds per patient).
Results: B-spline-based registration with intensity correction and MSE is compared against three different
methods: rigid using MI, B-splines using gradient magnitude (GM), both without preprocessing, and B-splines using MSE with global intensity adjustment. Overall, the proposed approach performs better than the other methods in terms of Dice similarity and mean surface distance of propagated contours. Mean Dice for the prostate is 0.79 while the second-best method (B-spline, global intensity adjustment) achieves 0.77. In two cases, none of the deformable methods produce better scores than rigid registration, which is likely due to poor image quality.
Conclusion: We introduce a patch-based intensity-correction approach that combats local inconsistencies in CBCT intensities. The method can be utilized as a preprocessing step for deformable registration and improves registration results, although the sample size in this study is too small to report statistical significance.