View on GitHub

Nifty Assignment: Training a neural network for medical image processing


Medical imaging is becoming an increasingly popular application of Machine Learning. Medical practitioners can use software to obtain diagnosis or second opinions on X-Ray images, lowering the chances of a missed threat.

In this assignment, students will be able to set up a model to train using the Deep Learning Tool Kit and Tensorflow.


Attribute Explanation
Summary Set-up an environment with TensorFlow and the Deep Learning Tool Kit to train and run inference on sample medical imagesets.
Topics Machine Learning, Neural Networks, TensorFlow.
Audience CS0, CS1 to get their feet wet and introduce concepts, CS2+ for development.
Difficulty Easy to set up and get running, with little room for error.
Strengths The strength of this assignment is being able to create a training environment quickly.
Weaknesses The topics that the Deep Learning Tool Kit and TensorFlow require are hard to digest and understand.
Dependencies Requires understanding of Machine Learning training models and understanding of visual data.
Variants Advanced Machine Learning classes may delve into the source code of the training models.

System Requirements

Assignment instructions

Setting up DLTK

  1. Clone the DLTK git repository
    git clone
  2. Navigate to the DLTK folder
    cd DLTK
  3. Create a virtual python environment
    virtualenv -p python2 venv_tf
  4. Activate the virtual environment
    source venv_tf/bin/activate
  5. Install the dependencies for DLTK within the virtual environment
    pip install -e .
  6. Install tensorflow
    pip install tensorflow-gpu==1.4.0

Downloading the IXI data sets for training

For more information on the IXI dataset, visit

  1. Go to the IXI_HH folder under data
    cd DLTK/data/IXI_HH
  2. Run the download script

    Note that the download may take a long time and may need to be left overnight depending on your internet connetion.

Start training

IXI_HH sex classification using resnet

Note: Any time you are training or inferencing, make sure the python virtual environment is enabled

   source DLTK/venv_tf/bin/activate
  1. Go to the IXI_HH_sex_classification folder
    cd DLTK/examples/applications/IXI_HH_sex_classification
  2. Create a folder to keep your model progress in
    mkdir my_model
  3. Begin training
    python --verbose --cuda_devices 0 --model_path my_model
  4. Start up TensorBoard to observe the training progress
    tensorboard --logdir .
  5. Open a browser to access Tensorboard, by default, Tensorboard opens on port 6006

Tensorboard interface

Accuracy and Loss graphs

Accurcy and Loss

Image evaluation


t-SNE graphs

For information on how to effectively understand t-SNE graphs, see the following Distill entry:



By default, inference runs on random images from the data set and infers whether they are male or female.

   python -u --model_path my_model



Assignment possible thanks to the work of the great people at the Deep Learning Tool Kit