Article
Implementation of an elastic registration algorithm for comparison of neuroradiological three-dimensional image data and angiograms
Implementierung eines elastischen Registrierungsalgorithmus zum Vergleich komplexer dreidimensionaler Datensätze
Search Medline for
Authors
Published: | April 11, 2007 |
---|
Outline
Text
Objective: For comparison of 3D angiographic data sets obtained from different imaging sources, such as rotational digital subtraction angiography (rDSA) and magnetic resonance angiography (MRA), a registration algorithm is needed that is able to detect local changes and displacements of structures. MRA and rDSA are geometrically not congruent. Therefore, simple visual comparison or 2D measurements are intrinsically afflicted with varying magnitude of error.
Methods: We developed a module for elastic registration of two data sets and integrated this module into the commercially available visualisation software system Amira (Mercury Computer Systems GmbH Berlin; TGS Inc. San Diego). Using free software libraries of the Insight Segmentation and Registration Toolkit (www.itk.org) a deformable registration algorithm was implemented. For registration a deformable grid is overlaid over one data set. The grid has a free defined number of grid nodes. The deformation of the data set is done by the local movement of each grid node. The new position of the voxels in the deformed data set between the grid node positions are interpolated with a parameterised cubic curve (B-spline). The movement of the grid nodes are performed with respect to an optimisation of the similarity between the deformed data and the reference data. Since we want to register multimodal data sets we used a similarity measure based on the mutual information. Due to the large number of parameters (each grid node can move along the 3 axes) a wide variety of deformations can be computed. However, this also results in high computation time. As result of the registration process a new data set is generated that can be saved separately.
Results: Previous to elastic registration a rigid registration had to be performed in order to align the structures in principle. Even though we use a standard Windows based computer the elastic registration process is done within an acceptable time of 2 to 5 minutes. The resolution of some data sets had to be reduced because of the limited working memory. Registered data sets show a better alignment of structures.
Conclusions: With the elastic registration a better comparison of follow-up examinations can be achieved. The visualisation of the local differences makes it easier to identify changes over the time. Therefore, implementation of this algorithm holds promise to offer immediate impact on clinical therapeutic decisions and choice of intervention.