Multi-Modal MRI Reconstruction Assisted with Spatial Alignment Network

Abstract

In clinical practice, multi-modal magnetic resonance imaging (MRI) with different contrasts is usually acquired in a single study to assess different properties of the same region of interest in human body. The whole acquisition process can be accelerated by having one or more modalities under-sampled in the $k$-space. Recent researches demonstrate that, considering the redundancy between different modalities, a target MRI modality under-sampled in the $k$-space can be more efficiently reconstructed with a fully-sampled reference MRI modality. However, we find that the performance of the aforementioned multi-modal reconstruction can be negatively affected by subtle spatial misalignment between different modalities, which is actually common in clinical practice. In this paper, we improve the quality of multi-modal reconstruction by compensating for such spatial misalignment with a spatial alignment network. First, our spatial alignment network estimates the displacement between the fully-sampled reference and the under-sampled target images, and warps the reference image accordingly. Then, the aligned fully-sampled reference image joins the multi-modal reconstruction of the under-sampled target image. Also, considering the contrast difference between the target and reference images, we have designed a cross-modality-synthesis-based registration loss in combination with the reconstruction loss, to jointly train the spatial alignment network and the reconstruction network. The experiments on both clinical MRI and multi-coil $k$-space raw data demonstrate the superiority and robustness of the multi-modal MRI reconstruction empowered with our spatial alignment network. Our code is publicly available at https://github.com/woxuankai/SpatialAlignmentNetwork.

Overview

The above figure is a real case demonstrating the existence of spatial misalignment (a), and the overview of the proposed method (b). In (a), a real case of multi-modal MRI acquired for the diagnostic purpose demonstrates the existence of spatial misalignment (highlighted by arrows) between the reference (T1-weighted) and the target (T2-weighted) images. The aligned reference image is also available to show the effect of our proposed spatial alignment network. In (b), a spatial alignment network is integrated into the multi-modal MRI reconstruction pipeline to compensate for the spatial misalignment between the fully-sampled reference image and the under-sampled target. The data flow for the conventional deep-learning-based reconstruction is shown in black arrows; and the red arrows are for additional data flow related to our proposed spatial alignment network.

For more details on the proposed method, please refer to https://ieeexplore.ieee.org/document/9745968.

Experiments on fastMRI DICOM

Prepare data

Store data in h5 files

1. Unzip fastMRI brain DICOM to fastMRI_brain_DICOM folder.
2. Convert all dicom to brain_nii folder.

ls fastMRI_brain_DICOM | while read X; \
do XX="brain_nii/${X}"; mkdir ${XX}; \
echo "dcm2niix -z n -f '%j-%p' -o ${XX} fastMRI_brain_DICOM/${X} 2>${XX}/error.log 1>${XX}/out.log"; \
done | parallel --bar

3. Convert selected nii to h5.

# T1 weighted
cat t1_t2_paired_6875_{train,val,test}.csv | cut -f1 -d ',' | while read x; \
do python3 convert_fastMRIDICOM.py "${x%.h5}.nii" "${x}" T1; \
done
# T2 weighted
cat t1_t2_paired_6875_{train,val,test}.csv | cut -f2 -d ',' | while read x; \
do python3 convert_fastMRIDICOM.py "${x%.h5}.nii" "${x}" T2;
done

Training & Evaluation

Run commands_train_eval.sh to start training and evaluation of methods.

Cite This

@ARTICLE{9745968,
  author={Xuan, Kai and Xiang, Lei and Huang, Xiaoqian and Zhang, Lichi and Liao, Shu and Shen, Dinggang and Wang, Qian},
  journal={IEEE Transactions on Medical Imaging}, 
  title={Multi-Modal MRI Reconstruction Assisted with Spatial Alignment Network}, 
  year={2022},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TMI.2022.3164050}}