LSN

Longitudinal Siamese Network for clinical trajectory prediction

• Code modules

  • Trajectory prediction (can be used to predict other clinical tasks as well)

    • LSN/notebooks/LSN_sim_testcode.ipynb: Stand-alone notebook for testing LSN with simulated data
    • LSN/lib/lsn.py: LSN model class and useful functions for training and testing
    • LSN/notebooks/run_lsn.ipynb: notebook to train and test LSN model with real data

  • Trajectory modeling

    • LSN/notebooks/model_trajectories.ipynb: Clustering code for modeling longitudinal clinical trajectories and subsequent assignment to new subjects

• LSN data flow

  • input (check notebooks for required data shapes)
    • MR: baseline + follow-up (e.g. 78x2 AAL CT values)
    • aux: apoe4 status + clinical scores (baseline + follow-up) + demographics (optional)
  • output (one-hot)
    • labels: trajectory / Dx / Px labels (binary and multiclass are supported)

• Legacy dir has older code version along with useful notebook for mapping vertex-wise CIVET data into an atlas based ROIs.

• Prereqs

  • python3.5+
  • tensorflow-gpu 1.4.1 (conda: conda install -c anaconda tensorflow-gpu)
  • sklearn
  • pandas
  • seaborn