Quick Links

• Development Branch

• Experiments Branch

• Augmentations Notebook

Fetal Total-Body and Brain 3D Segmentation in MRI scans with limited datasets.

• 3D and 2D segmentation.
• Implemented Advanced Augmentation techniques.
• Achieving results below Radiologist Inter & Intra Observer variability.
• Option to Utilize previous-slice segmentation to improve performance on hard samples.

Training

• Install Tensorflow & Keras

• Install dependencies:
  pip install -r requirements.txt

• To run training:
  Config currently in dict inside brats/train_fetal.py

$ python -m brats.train_fetal --config_dir <config_dir>

Where <config_dir> is the folder containing the training configurations (or will be creatad to)</config_dir>

Write prediction images from the validation data

In the training above, part of the data was held out for validation purposes.
To write the predicted label maps to file:

$ python -m brats.predict --config_dir <config_dir>

The predictions will be written in the <config_dir>/prediction folder along with the input data and ground truth labels for comparison.

More Instructions and Explanations in the Wiki!

Cool Results

3D Model of the segmented fetal brain

3D Model of the segmented fetal body

Segmentation of fetal brain

Segmentation of fetal brain (only border)

Segmentation of fetal body