[COPD-MICCAI 2018] Try different architecture changes to Sub2Vec

[sumedhasingla]

• [ ] Try aleast 4 different methods to include expiratory images.

• [ ] Extend Auto encoder decoder to have skip connections like in UNet

[kayhan-batmanghelich]

@sumedhasingla today I realized that we don't have skipped connection. Also, another thing that was pointed out by some that there is information outside of the lung region. Not for this journal but for other genetic prediction, we might need to change the architecture completely. Here is an architecture that can work:

We can apply a method that is adaptable with input size for a subset of slices and then combine it with LSTM to embed the entire volume. This way we analyze the entire volume, look outside the lung, don't lose spatial info, and handle memory issue.