Fourier-Net: Fast Image Registration with Band-Limited Deformation

Authors

  • Xi Jia School of Computer Science, University of Birmingham, UK
  • Joseph Bartlett School of Computer Science, University of Birmingham, UK Department of Biomedical Engineering, University of Melbourne, Australia
  • Wei Chen School of Computer Science, University of Birmingham, UK
  • Siyang Song Department of Computer Science and Technology, University of Cambridge, UK
  • Tianyang Zhang School of Computer Science, University of Birmingham, UK
  • Xinxing Cheng School of Computer Science, University of Birmingham, UK
  • Wenqi Lu Department of Computer Science, University of Warwick, UK
  • Zhaowen Qiu Institute of Information Computer Engineering, Northeast Forestry University, China
  • Jinming Duan School of Computer Science, University of Birmingham, UK Alan Turing Institute, UK

DOI:

https://doi.org/10.1609/aaai.v37i1.25182

Keywords:

CV: Medical and Biological Imaging

Abstract

Unsupervised image registration commonly adopts U-Net style networks to predict dense displacement fields in the full-resolution spatial domain. For high-resolution volumetric image data, this process is however resource-intensive and time-consuming. To tackle this problem, we propose the Fourier-Net, replacing the expansive path in a U-Net style network with a parameter-free model-driven decoder. Specifically, instead of our Fourier-Net learning to output a full-resolution displacement field in the spatial domain, we learn its low-dimensional representation in a band-limited Fourier domain. This representation is then decoded by our devised model-driven decoder (consisting of a zero padding layer and an inverse discrete Fourier transform layer) to the dense, full-resolution displacement field in the spatial domain. These changes allow our unsupervised Fourier-Net to contain fewer parameters and computational operations, resulting in faster inference speeds. Fourier-Net is then evaluated on two public 3D brain datasets against various state-of-the-art approaches. For example, when compared to a recent transformer-based method, named TransMorph, our Fourier-Net, which only uses 2.2% of its parameters and 6.66% of the multiply-add operations, achieves a 0.5% higher Dice score and an 11.48 times faster inference speed. Code is available at https://github.com/xi-jia/Fourier-Net.

Downloads

Published

2023-06-26

How to Cite

Jia, X., Bartlett, J., Chen, W., Song, S., Zhang, T., Cheng, X., Lu, W., Qiu, Z., & Duan, J. (2023). Fourier-Net: Fast Image Registration with Band-Limited Deformation. Proceedings of the AAAI Conference on Artificial Intelligence, 37(1), 1015-1023. https://doi.org/10.1609/aaai.v37i1.25182

Issue

Section

AAAI Technical Track on Computer Vision I