HDAR: Hierarchical Difficulty-Aware for deformable Registration

HDAR Framework

We present a difficulty-aware model based on an attention mechanism to automatically identify hard-to-register regions, allowing better estimation of large complex deformations. The difficulty-aware model is incorporated into a cascaded neural network consisting of three sub-networks to fully leverage both global and local contextual information for effective registration. Embedding difficulty-aware learning into the hierarchical neural network allows harder patches to be identified in the deeper sub-networks at higher resolutions for refining the deformation field. Please refer to our paper for more details.

Difficulty-aware Patch Selection

Dataset

1. LONI40
2. IBSR18
3. CUMC12
4. MGH10

Comparsion with State-of-the-Art Methods

Example registration results given by the D. Demons, LCC-Demons, SyN, and HDAR.

Comparisons with VoxelMorph

Installation

This code requires Tensorflow-GPU 1.14, TensorLayer 1.10 and Python 3.6.

Citation

If you use this code for your research, please cite our paper.

@article{HUANG2021101817,
title = "Difficulty-aware hierarchical convolutional neural networks for deformable registration of brain MR images",
journal = "Medical Image Analysis",
volume = "67",
pages = "101817",
year = "2021",
issn = "1361-8415",
doi = "https://doi.org/10.1016/j.media.2020.101817",
author = "Yunzhi Huang and Sahar Ahmad and Jingfan Fan and Dinggang Shen and Pew-Thian Yap"
}

Acknowledgments

The source code is inspired by VoxelMorph