Premium
SU‐F‐J‐96: Comparison of Frame‐Based and Mutual Information Registration Techniques for CT and MR Image Sets
Author(s) -
Popple R,
Bredel M,
Brezovich I,
Dobelbower M,
Fisher W,
Fiveash J,
Guthrie B,
Riley K,
Wu X
Publication year - 2016
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4956004
Subject(s) - mutual information , fiducial marker , image registration , nuclear medicine , computer vision , artificial intelligence , frame (networking) , computer science , medical imaging , mathematics , medicine , image (mathematics) , telecommunications
Purpose: To compare the accuracy of CT‐MR registration using a mutual information method with registration using a frame‐based localizer box. Methods: Ten patients having the Leksell head frame and scanned with a modality specific localizer box were imported into the treatment planning system. The fiducial rods of the localizer box were contoured on both the MR and CT scans. The skull was contoured on the CT images. The MR and CT images were registered by two methods. The frame‐based method used the transformation that minimized the mean square distance of the centroids of the contours of the fiducial rods from a mathematical model of the localizer. The mutual information method used automated image registration tools in the TPS and was restricted to a volume‐of‐interest defined by the skull contours with a 5 mm margin. For each case, the two registrations were adjusted by two evaluation teams, each comprised of an experienced radiation oncologist and neurosurgeon, to optimize alignment in the region of the brainstem. The teams were blinded to the registration method. Results: The mean adjustment was 0.4 mm (range 0 to 2 mm) and 0.2 mm (range 0 to 1 mm) for the frame and mutual information methods, respectively. The median difference between the frame and mutual information registrations was 0.3 mm, but was not statistically significant using the Wilcoxon signed rank test (p=0.37). Conclusion: The difference between frame and mutual information registration techniques was neither statistically significant nor, for most applications, clinically important. These results suggest that mutual information is equivalent to frame‐based image registration for radiosurgery. Work is ongoing to add additional evaluators and to assess the differences between evaluators.