https://github.com/kerkelae/replication-gyori-et-al

Code for replicating the paper "Training data distribution significantly impacts the estimation of tissue microstructure with machine learning"
https://github.com/kerkelae/replication-gyori-et-al

Last synced: about 9 hours ago
JSON representation

Code for replicating the paper "Training data distribution significantly impacts the estimation of tissue microstructure with machine learning"

• Host: GitHub
• URL: https://github.com/kerkelae/replication-gyori-et-al
• Owner: kerkelae
• License: mit
• Created: 2023-12-04T18:58:24.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2024-01-12T14:05:43.000Z (about 1 year ago)
• Last Synced: 2024-01-13T03:59:03.278Z (about 1 year ago)
• Language: Jupyter Notebook
• Homepage:
• Size: 696 KB
• Stars: 2
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
> 🚨 The data is no longer available to download.

# Replication of "Training data distribution significantly impacts the estimation of tissue microstructure with machine learning"

This repository contains the code for replicating the main results of the following paper:

Gyori, Noemi G., et al. _"Training data distribution significantly impacts the estimation of tissue microstructure with machine learning."_ Magnetic resonance in medicine 87.2 (2022): 932-947. [DOI:10.1002/mrm.29014]( https://doi.org/10.1002/mrm.29014)

## Installation

All the code in this repository has been run using [Anaconda](https://www.anaconda.com/download). The development environment with all the required packages is defined in `environment.yml`, which can be installed by executing

```

conda env create -f environment.yml

```

Then, you can activate the environment with

```

conda activate replication_env

```

## Download data

We will use a 2-shell high-angular resolution diffusion imaging (HARDI) data. You can download the preprocessed data or the raw data and run the preprocessing script yourself (this can take some time).

### Preprocessed data

Preprocessed data can be downloaded by executing

```

python fetch_preprocessed_data.py

```

### Raw data

Raw data can be downloaded by executing

```

python fetch_raw_data.py

```

Raw diffusion MRI data needs to be pre-processed before analysis. Running the preprocessing script requires [fsl](https://fsl.fmrib.ox.ac.uk/) to be installed. I ran the preprocessing using fsl version 6.0.7.3. You can run the preprocessing script by executing

```

bash preprocessing.sh

```

## Conventional model fit

In order to run the conventional model fit, you need to install [smt](https://github.com/ekaden/smt) and run

```

bash fit_smt.sh

```

## Training and comparison

Once the data has been prepared and the parameters estimated using the conventional model fit, the network can be trained using the notebooks `train_uniform.ipynb` and `train_healthy_brain.ipynb`. The results can be compared using the notebook `comparison.ipynb`.