SMuK 2023 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
ST: Fachverband Strahlen- und Medizinphysik
ST 8: AI Topical Day – AI in Medicine (joint session ST/AKPIK)
ST 8.1: Talk
Thursday, March 23, 2023, 14:00–14:15, ZEU/0146
Multimodal image registration with deep learning — •Alexander Ratke1, Christian Bäumer2, Kevin Kröninger1, and Bernhard Spaan1 — 1TU Dortmund University, Dortmund, Germany — 2West German Proton Therapy Centre Essen, Essen, Germany
In radiation therapy, precise localisation of tumour and risk structures is important for treatment planning. Medical imaging methods, such as computed tomography (CT) and magnetic resonance imaging (MRI), allow a differentiation between these structures. Planning systems typically align CT and MRI scans rigidly to compensate inaccurate immobilisation of the patient, but distortions in MRI or movement of organs still remain.
In this project, a data set of CT and MRI scans of the head and neck areas is used to study unsupervised deformable image registration with deep learning. First, the scans are pre-processed, which includes rigid registrations and the equalisation of the image formats. Then, deep learning is employed to filter structures of an image through multiple layers and to match them to a second image. The registration model strongly depends on the choice of its parameters. Therefore, variations of these parameters are investigated on the data set. The results are presented as well as the overall workflow including the pre-processing.