This repository provides the official implementation of the paper "FractMorph: A Fractional Fourier-Based Multi-Domain Transformer for Deformable Image Registration".
[Graphical Abstract] [Paper] [Code]
An overview of the FractMorph framework is shown below. Given a moving and a fixed 3D image, the model estimates the corresponding 4D deformation field. By novel integration of the fractional Fourier transform with a dual-parallel Transformer, the architecture simultaneously captures local, semi-global, and global context within a single end-to-end network.
The figure below presents the architecture of the multi-domain Fractional Cross-Attention (FCA) module.
The images below display the magnitude and phase outputs of our implemented 3D FrFT applied to cardiac (top two rows) and cerebral (bottom two rows) medical images at different fractional orders. For illustration, only a single 2D slice from each volume is shown.
The implementation of the 3D FrFT can be found in modules/frft3d.py.
If you find this code useful in your work, please consider citing:
@article{kebriti2025fractmorph,
title={FractMorph: A Fractional Fourier-Based Multi-Domain Transformer for Deformable Image Registration},
author={Kebriti, Shayan and Nabavi, Shahabedin and Gooya, Ali},
journal={arXiv preprint arXiv:2508.12445},
year={2025}
}[08/23/2025]The FrFT implementation became faster and more efficient.[08/19/2025]The preprint became publicly available.