Adaptive non-rigid registration of 3D knee MRI in different pose spaces

Abstract

Volume registration of articulated subjects such as the human knee region requires novel methods to accommodate the wide range of movement resulting from skeletal joint rotations. This paper addresses articulated volume deformation with a pair of techniques. Volume skeletal deformation (VSD) is applied first, producing a smooth but approximate deformation of the soft tissue as the underlying skeleton moves, while preserving rigid motion in the segmented bone regions. The resulting deformed volume is used as initialization for a nonrigid deformation step that locally and adaptively adds radial basis function control points to minimize the remaining registration error in any poorly aligned region. The resulting process registers a neutral pose volume to any given bent-knee pose MRI scan.

Scan-Based Volume Animation Driven by Locally Adaptive Articulated Registrations

“…the total DOFs can be reduced from 7000 to 24). It should be noted that the human hand MRI volume could not be registered using a global registration approach [33] due to strong local minima and dramatically increasing computation time with additional control points; our registration tests with this method failed after running for more than 72 hours. To our knowledge, our locally based approach is the first achievement to register entire human hand MRI volumes in different poses (including soft-tissue deformations) within a reasonable time.…”

“…In recent work by [33], human knee MRI volumes in highly different poses are adaptively registered. The volumetric linear blending algorithm [11] was used for initialization in combination with successive adaptive registration using global deformation functions such as RBFs based on TPS [28].…”

“…Compared with previous algorithms using global deformation [33], fewer than one percent of the total DOFs are required (e.g. the total DOFs can be reduced from 7000 to 24).…”

“…General non-rigid volume registration methods can be defined by a cost function minimizing the mismatch between the source and target volumes [33], max…”

“…Global deformation functions such as RBFs based on TPS [28], [33] cannot provide locally constrained deformation and therefore require an excessive number of control points and DOFs to register complex articulated subject such as the human hand. As shown in Fig.…”