https://github.com/hasibzunair/res-unet-fastmri

Last place solutioin to fastMRI Image Reconstruction Challenge 2019 (Single coil track).
https://github.com/hasibzunair/res-unet-fastmri

deep-learning mri-reconstruction super-resolution

Last synced: 4 days ago
JSON representation

Last place solutioin to fastMRI Image Reconstruction Challenge 2019 (Single coil track).

• Host: GitHub
• URL: https://github.com/hasibzunair/res-unet-fastmri
• Owner: hasibzunair
• Created: 2019-07-17T23:03:32.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2022-12-08T01:23:37.000Z (almost 2 years ago)
• Last Synced: 2023-04-13T07:20:44.127Z (over 1 year ago)
• Topics: deep-learning, mri-reconstruction, super-resolution
• Language: Jupyter Notebook
• Homepage: https://fastmri.org/leaderboards/challenge/2019/
• Size: 19.3 MB
• Stars: 8
• Watchers: 1
• Forks: 3
• Open Issues: 6
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
## fastMRI Image Reconstruction Challenge 2019 (Single-coil track)

![](media/mri_low.gif) | ![](media/mri_high.gif)

The project is structured as follows.

### Challenge description

Given an undersampled knee MRI scan, the goal is to reconstruct a high resolution knee MRI scan. More details about the dataset and task can be found [here](https://fastmri.org/dataset/).

### Our method

We processed the data at the slice level. For each knee MRI low resolution, there was a corresponding high resolution knee MRI. On this processed data, we trained a U-Net architecture with a pretrained ResNet backbone on the knee MRI slices. Refer to [this](https://github.com/hasibzunair/MRI-reconstruction/blob/master/unet.ipynb) notebook for code implementation.

### Dependencies

This work is implemented in Python 3.6 and Keras using Tensorflow as backend.

*    Ubuntu 14.04

*    Python 3.6

### Directory strucuture and usage

* `media` : Contains supporting material for README.md

* `dataset` : training data provided by competition

* `fastMRI` : fastMRI github repository for helpers and utils

* *.ipynb # notebooks and python scripts

* *.py

### Dataset directory strucuture:

```

dataset/

    singlecoil_train/

                # *.h5 files of MRI data

    singlecoil_test_v2/

                # *h5 raw test samples

    # preprocessed

    singlecoil_train_3D_images_48x/

                                low/

                                    # undersampled 3D image volumes

                                high/

                                    # ground truth 3D  image volumes

                           

```

### Challenge Leaderboard 2019

A total of 17 teams came into the final leaderboard, among which we were the last! Some logs are shown below.







### Reference to other models

Some helper scripts are based on https://github.com/facebookresearch/fastMRI.

### License

Your driver's license.