1/10 I am happy to present the first release of #NiftyTorch, a Python API for deploying #DeepLearning for #neuroimaging research. We tried to make it an easy-to-use library for the end-users. I hope neuroimaging community benefit from this as we really worked hard to get here.

2/10 full documentation is here: https://niftytorch.github.io/doc/ . On this thread I highlight some of the key features of NiftyTorch, starting with: intuitive and simple modules.

3/10 #PyTorch Embedded end-to-end data-loading pipeline. NiftyTorch has image-to-label and image-to-image data loaders, reducing data handling difficulties.

4/10 Built-in Attention module for incorporating demographic data. Users can plug-in #demographic, #clinical or #behavioral files (e.g. meta data csv files) into their #CNN networks.

5/10 Automated hyperparameter optimization. NiftyTorch has trainer modules that explore solution space using #Optuna library for automated hyperparamter #optimization

6/10 Several loss function (some not included in PyTorch). Simple interaction with GPU and multi-scaled training (e.g. data parallelization).

7/10 Easily customize your own network. Combine different #DeepLearing networks and create your own.

8/10 All the convolutions and networks has been designed to work with #3D and large-scale images and #bigdata. No 2D shortcut has been made.

9/10 Getting started and basic demos are included here: https://github.com/NiftyTorch/NiftyTorch.v.0.1/tree/master/Demo. We also included more advanced demos such as network optimization and customization. and a Unet demo

10/10 Please follow NiftyTorch @NiftyTorch for news and updates. Please give us constructive feedback. And Please support and acknowledge this effort. And please retweet. Thank you for reading this thread.