Developing Brain Atlas through Deep Learning

SeBRe (Segmenting Brain Regions) is a high-throughput Mask R-CNN [1]-based toolbox to generate brain atlas through deep learning.

Iqbal, A., Khan, R. and Karayannis, T., 2019. Developing a brain atlas through deep learning. Nature Machine Intelligence, 1(6), p.277.

Iqbal, Asim, Romesa Khan, and Theofanis Karayannis. "Developing Brain Atlas through Deep Learning." arXiv preprint arXiv:1807.03440 (2018).

Download:

Clone the SeBRe repository by running the following command in your terminal window:

git clone https://github.com/itsasimiqbal/SeBRe.git

Additionally, download the (.zip) files by visiting the following link:

https://drive.google.com/open?id=1g_894zM_mSJGfudzzfnw-ZpN-rQ4qQdz

You would need DATASETSubmit.zip, myDATASET.zip and SeBRe_FINAL_WEIGHTS.h5.zip to run the code in your machine. If you'd like to use the mouse and human brain atlas datasets used in our paper then also download and unzip SeBRe_Datasets.zip. Once you've downloaded your files then unzip them in your folder by running the following commands:

unzip DATASETSubmit.zip
unzip myDATASET.zip
unzip SeBRe_FINAL_WEIGHTS.h5.zip

SeBRe demo:

The block diagram of our system is demonstrated below:

Run the notebook SeBRe_FINAL.ipynb to reproduce the results in SeBRe's paper. Make sure to install the necessary libraries in your machine before running the code. A step-by-step explanation of feature processing in SeBRe is provided in SeBRe_feature_processing.ipynb notebook.

To train SeBRe on your (custom) dataset:

1. Collect images of brain regions (or sub-regions) and place each of them in a separate folder with the following naming convention e.g. section_img_0.jpg, section_img_1.jpg, ...

2. Draw ground-truth (binary) masks on Regions of Interest (ROIs) e.g. cortex, hippocampus, etc. using a scalar vector graphics (SVG) software such as Boxy SVG [2], RectLabel [3], Inkscape [4], etc. and assign a unique color code to each ROI. In the figure below, mouse brain section is shown before (a) and after (d) annotation by human expert. A zoomed-in examples of cortex (b) and hindbrain (c) are shown to precisely match the boundaries of masks with the corresponding ROIs.

3. Run the notebook custom_dataset_create.ipynb to generate the binary (black and white) masks to train SeBRe deep neural network (DNN). The notebook will generate the file names in SeBRe-readable format, place them in the brain region corresponding folders e.g. section_masks_0/section_masks_0_m_1.png, sections_masks_0_m_2.png, ...

4. Modify and run the notebook SeBRe_training.ipynb to train the SeBRe DNN on your (custom) dataset.

5. Modify and run the notebook SeBRe_FINAL.ipynb to test the SeBRe DNN on your (custom) dataset.

References:

[1] https://arxiv.org/abs/1703.06870

[2] https://boxy-svg.com/

[3] https://rectlabel.com/

[4] https://inkscape.org/