Ladislav Bari
Supervisor(s): Ing. Lukáš Hudec
Slovak Technical University
Abstract: Image registration with deep learning methods has become an active field of research and an exciting area for long-standing problems in medical imaging. However, supervised learning requires a large amount of accurately annotated corresponding control points, which can be difficult to obtain in the medical imaging domain. Unsupervised learning provides us an option to get rid of manual annotation by exploiting unlabeled data without supervision. The goal is to optimize a neural network to map the appearance of input image pairs to parameters of a spatial transformation to align corresponding anatomical structures, in our case MRI brain scans of patients suffering from Alzheimer’s disease. In this paper, we propose a method based on the usage of the Spatial transformer network which proved applicability for the registration task. We evaluate our method on 3D medical images from the TADPOLE dataset. We demonstrate that with the usage of affine transformations, our method outperforms the classical methods such as various registration methods provided in SimpleITK library.
Keywords: Computer Vision
Full text: Year: 2020
