Research project

Feature Point Selection and Motion Models for 2-D/3-D Registration

In interventional radiology, live X-ray images are used in order to guide the surgeon during the procedure. However, important anatomical structures may not be visible in these images. To visualize these structures, pre-operatively acquired 3-D images, such as CT- or MRI- scans, can be overlaid with the 2-D image. 2-D/3-D registration methods are used in order to estimate the pose of the 3-D image for the overlay. A feature-based registration method has been developed at the LME which can cope especially well with registration using a single 2-D image. This method makes use of a motion model which is able to estimate rigid 3-D transformations from 2-D displacements of a set of points.
In this research project, feature point selection methods and motion model extensions are explored which can further improve and extend the robustness as well as the accuracy of the registration method. Although different feature extraction methods are described in the literature, the setting in medical 2-D/3-D registration is unique in that the imaged objects are translucent for the X-ray imaging system. The selection of good points is done dependent on the use case. This includes feature selection depending on the anatomical structures which are registered as well as multi-modal registration, where the feature-matching is more challenging due to different intensity distributions for the same anatomical structures. Extensions of the motion model are also considered to make optimal use of the displacement information which can be gained depending on the feature properties (for example, only a 1-D component of the displacement can be estimated at edge points due to the aperture problem while 2-D displacement can be estimated at corners).

Project manager:
Prof. Dr.-Ing. habil. Andreas Maier

Project participants:
Roman Schaffert, M. Sc., Jian Wang, M. Sc., Dr.-Ing. Anja Borsdorf

Keywords:
Rigid; Registration; Fluoroscopy; CT; Feature-Based; 2-D/3-D; Motion Estimation;

Duration: 1.8.2015 - 31.7.2019

Sponsored by:
Siemens Healthineers AG

Contact:
Schaffert, Roman
Phone +49 9131 85 27826, Fax +49 9131 85 27270, E-Mail: roman.schrom.schaffert@fau.de

Publications

Schaffert, Roman ; Wang, Jian ; Fischer, Peter ; Borsdorf, Anja ; Maier, Andreas: Multi-View Depth-Aware Rigid 2-D/3-D Registration. In: IEEE (Ed.) : 2017 IEEE Nuclear Science Symposium and Medical Imaging Conference Record (NSS/MIC) (IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC) Atlanta, Georgia, USA 26.10.2017). 2017, pp -.

Schaffert, Roman ; Wang, Jian ; Borsdorf, Anja ; Hornegger, Joachim ; Maier, Andreas: Comparison of Rigid Gradient-Based 2D/3D Registration Using Projection and Back-Projection Strategies. In: Tolxdorff, Thomas ; Deserno, M. Thomas ; Handels, Heinz ; Meinzer, Hans-Peter (Ed.) : Bildverarbeitung für die Medizin (Workshop Bildverarbeitung für die Medizin Berlin 13.03.2016). Springer : Springer, 2016, pp 140-145.
[doi>10.1007/978-3-662-49465-3_26]

Schaffert, Roman ; Wang, Jian ; Fischer, Peter ; Borsdorf, Anja ; Maier, Andreas: Metric-Driven Learning of Correspondence Weighting for 2-D/3-D Image Registration. In: Springer (Ed.) : Pattern Recognition, 40th German Conference (40th German Conference on Pattern Recognition (GCPR 2018) Stuttgart 10.10.2018-12.10.2018). 2018, pp 1-13.