DeepEnReg: Joint Enhancement and Affine Registration for Low-contrast Medical Images

Xiaolin Wang, Ziyang Li, Yun Peng, Di Hu, Huanyu Luo, Huanjie Li, Zhongxuan Luo, Xin Fan

Abstract

Affine registration aims to find the low-dimensional parametric transformation that best aligns one data to another. However, existing registration methods, either classic energy optimization or deep learning are mainly designed for adult brain images and have limited performance on infant brain images with widely varied intensity distributions and low-contrast issues. To achieve fast and robust registration on low-contrast infant brain images, we propose an unsupervised deep registration framework DeepEnReg with a deep enhancement module and a deep affine registration module. Our affine registration module leverages a multi-resolution loss to guarantee consistency on sparsely sampled infant brain images. Our DeepEnReg achieves reasonable and reliable performance on the affine registration tasks of infant brain images and synthetic data and significantly reduces irregular registration results compared to other mainstream methods. Our proposed method significantly improves the computation efficiency over the mainstream medical image processing tools (from 13 to 0.570 s for a 3D image pair on affine registration) and outperforms state-of-the-art approaches.
BibTeX
@inproceedings{wang2022deepenreg,
  author = {Xiaolin Wang and Ziyang Li and Zi Li and Yun Peng and Di Hu and Huanyu Luo and Huanjie Li and Zhongxuan Luo and Xin Fan},
  title = {{DeepEnReg: Joint Enhancement and Affine Registration for Low-contrast Medical Images}},
  booktitle = {PRCV},
  year = {2022},
  doi = {10.1007/978-3-031-18910-4\_13},
}

Loading related content...

Figure