7T MRI Synthesization from 3T Acquisitions

🔗 View paper on arXiv

This repository contains PyTorch model implementations for the paper "7T MRI Synthesization from 3T Acquisitions", presented at MICCAI 2024 conference. This model generates synthetic T1-weighted 7T MRIs from T1-weighted 3T MRI inputs. Implemented models are V-Net, Perceptual V-Net, V-Net-GAN, WATNet-2D, and WATNet-3D.

Requirements

To install requirements:

pip install -r requirements.txt

Usage

Example command:

        python run_vnet.py -i 'path_to_input (single file or folder with files)'  -o 'folder_to_save_output' -c 'path_to_model_weight'

• Please make sure the input images meet the following criteria:
  • Nifti file
  • T1-weighted 3T image
  • Brain stripped (provide mask files as an optional input under -m if the image is not brain stripped.)

Training

• The training parameters should be specified in config/params.json.
• Place paired dataset in data/,and update the dataset filepaths in config/params.json.
• To initiate training, run src/scripts/Run_model.py

Data Augmentation

• The script for data augmentation is src/scripts/data_augmentation.ipynb
• Transformed datasets are saved under data/

Pretrained Weights

• Pretrained weights for the base V-Net model can be found on 🔗 Box

Paper BibTex Citation

If you use this tool, please cite the following reference:

@InProceedings{Cui_7T_MICCAI2024,
        author = { Cui, Qiming and Tosun, Duygu and Mukherjee, Pratik and Abbasi-Asl, Reza},
        title = { { 7T MRI Synthesization from 3T Acquisitions } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15007},
        month = {October},
        page = {35 -- 44}
}