BFRnet
March 20, 2026 · View on GitHub
BFRnet: A Deep Learning-Based MR Background Field Removal Method for QSM of the Brain Containing Significant Pathological Susceptibility Sources
arXiv | data & checkpoints | deepMRI collection
BFRnet performs background field removal for QSM, specifically designed to handle brains with significant pathological susceptibility sources (e.g., hemorrhage, calcification) where conventional methods struggle.
Overview
Requirements
- MATLAB R2019a+
- NVIDIA GPU (16 GB RAM recommended for demo)
Quick Start
1. Download pre-trained model and demo data
Download from Dropbox and place in the Eval/ folder.
2. Run demo
% Navigate to Eval folder and run:
BFRnet_demo
A human brain COSMOS map and total field map are included for testing. 16 GB RAM recommended.
Training
Generate training data
% Add the Train/ folder to your MATLAB path, then:
Gen_HighSus % generates synthetic background susceptibility and field maps
PhanGene % generates synthetic background susceptibility phantom
Train BFRnet
TrainBFRnet % GPU or HPC strongly recommended
Citation
@article{bfrnet2022,
title={BFRnet: A deep learning-based MR background field removal method for QSM of the brain containing significant pathological susceptibility sources},
journal={arXiv},
year={2022},
url={https://arxiv.org/abs/2204.02760}
}