Learn a Task-Adaptive MR Under-Sampling Pattern

A deep learning framework to learn a task-adaptive under-sampling pattern and reconstruct MRI in end-to-end way. This is a course project of BME 548 Machine Learning and Imaging at Duke University.

Get Started

Prerequisites

This project is based on python. Before starting playing with it, please ensure that you are familiar with python environment setup.

Installation

1. Clone the repo

git clone https://github.com/ZihaoChen0319/Deep-MR-Reconstruction-And-Undersampling-Pattern-Learning

2.Install the required python packages

pip install -r requirements.txt

Usage

Prepare the dataset

The datasets used here are from Medical Segmentation Decathlon, which contains 10 different medical image segmentation tasks. You can download the datasets by yourself or use your own datasets.

Please place the dataset at ./Data or the dir you like. The structure of dataset file folder should be like:

./Data
└── your_dataset
    ├── imagesTr
    ├── labelsTr
    └── (optional) dataset.json

Data preprocessing

Please set data_dir and data_name in data_preprocess.py based on your situation, then run:

python data_preprocess.py

The processed data will be saved at data_dir/data_name_np.

Training

Please set if_train=True and other parameters in main.py based on your situation, then run:

python main.py

Evaluation

Please set if_train=False in main.py and run:

python main.py

Roadmap

• Implementation of the framework
• Add a framework figure
• Use argparse in data_preprocess.py, main.py and visualize.py
• Enable the cartesian under-sampling pattern

Acknowledgement

Open-Source Repositories

• cagladbahadir/LOUPE
• NVIDIA/DeepLearningExamples

References:

• C. D. Bahadir, A. Q. Wang, A. V. Dalca, and M. R. Sabuncu, “Deep-Learning-Based Optimization of the Under-Sampling Pattern in MRI,” IEEE Transactions on Computational Imaging, vol. 6, pp. 1139–1152, 2020