Author: Zaccharie Ramzi, Francois Lanusse

Year: 2020

Email: zaccharie.ramzi@gmail.com, francois.lanusse@cea.fr

This tutorial was intended for the Covid-19 crisis learning-slots in March-April 2020. This intends to be a first introduction to TensorFlow 2.0 not as much from a machine learning point of view but from an optimization and inverse problem perspective.

We propose two notebooks:

• First Steps with TensorFlow: A short introduction to the basic concepts underpinning TensorFlow, in particular automatic differentation.

• MRI reconstruction with TensorFlow:
  An illustration of how to use TensorFlow in 2 different settings, for the same problem of MRI reconstruction:

  • as an optimisation library to perform classical iterative reconstruction
  • as a deep learning library to train a CNN to reconstruct fourier undersampled images

All of the notebooks can be run directly on Google Colaboratory, this is highly recommended.

If you want to run the tutorials locally, you will need Python 3.5+. All the Python requirements are listed in requirements.txt except cython which is to be installed beforehand. You can then install all the requirements using pip install -r requirements.txt.