Fig 1.
Synthetic experiments setup and results.
From left to right: Source, target and pullback-registered CC-DIR images for Linear (Top row), Nonlinear (Center row) and Growth (Bottom row) case. The registered images show clear matching of the shapes IL and IR.
Table 1.
Intersection (IS) and DICE score for synthetic experiments.
The purposed CC-DIR based algorithm shows no overlap in the three test cases additionally with a superior DICE score.
Table 2.
Comparison of TREs for the DIR-LAB data set in mm.
The mean TRE of the purposed CC-DIR method ranks at fourth place out of seven with an offset of 0.15 mm to the best performing algorithm.
Fig 2.
Overview of the reference, source and registered image.
|I − J| shows the intensity differences between moving and fixed image without registration, |I∘Φ − J| after collision-based registration. A clear reduction of intensity differences and matching of anatomical structures can be observed.
Fig 3.
Close-up of the displacement field.
Displacements around the interface (green arrows) show observable sliding motion along the boundary (red).
Table 3.
Comparison of gap/overlap for the DIR-LAB data set by case.
The purposed CC-DIR algorithm shows the least overlap compared to the six other methods. It rates second in terms of interface gap with a mean offset of 38.1 cm3 to the first place. All measurements are in cm3.
Fig 4.
High differences in shear stretches along lung-rib-cage interface with respect to the rest of the image are depicted.