Code for running deep learning on 2D projections from 3D volumes

Deep learning on large 3D medical volumes is very computationally demanding. We have demonstrated that using 2D projections from 3D volumes can lead to reasonable performance (for brain age prediction from MRI volumes), with a much lower computational complexity. We share Julia code for running 2D CNNs with projections as input channels.

The Julia code performs training of 2D convolutional neural networks (CNN) with different projections as input channels, to perform classification or regression.

The code is written in the Julia programming language

Anyone using deep learning on 3D volumes

See the GitHub repository