Room: Exhibit Hall | Forum 5
Purpose: This study aims at assessing the quality, accuracy, and plausibility of the Velocity AI deformable image registration in cases involving 4DCT scans based on various similarity measures.
Methods: 10 lung cancer patients were used. Their 4DCT scans were registered using 3 different DIR
methods: 1) RIGID â€“ translation and rotation in the x, y, and z planes); 2) deformable multi pass (DMP) â€“ a three pass coarse to medium-to-fine resolution deformable; and 3) extended deformable multi pass (EXDMP) â€“ a six pass deformable that goes into finer resolution than the DMP. The 3 DIR methods were analyzed using 24 similarity measures to evaluate image registration quality. To ensure the invertibility of the examined methods and assess the resultant mechanical stress, the determinant of the Jacobian for the displacement field and 3-D Eulerian strain tensor were calculated. The 3 DIR methods were evaluated using: a) the complete CT dataset; and b) the cropped CT dataset (3D CT dataset cropped to the tumor volume region).
Results: A similar trend was observed for the 3 DIR methods by several of the studied measures. The DMP method was always associated with positive Jacobian determinant values. The DMP method had smaller values than EXDMP in all of the studied cases regarding the Eulerian strain tensor evaluation. Additionally, the Eulerian strain tensor values indicate less tissue strain for the DMP method. The EXDMP method demonstrated non-physical behavior according to the negative values of the Jacobian determinant. Large differences were also observed in the results between complete and cropped datasets (coefficient of determination: 0.55 vs. 0.93).
Conclusion: The EXDMP method showed an overwhelming non-physical and unrealistic behavior as well as poor image similarity in a number of studied cases, making DMP the method of choice for the thoracic anatomical region.
Image Fusion, Deformation, Shape Deformation