SonadorAI: MVP

Determine use cases and requirements
- DataSet: Get data from Orthanc to DL frameworks
  - High variability of how data should be structured based on the dataset.
  - High variability on where label data is stored (private tags, DICOMseg files, annotations)
  - DataSet should provide a bridge from Sonador/Orthanc to PyTorch tensor. Look at TensorFlow/Keras in the near future.
- Data reproducibility: create tools to help manage ML lifecycle. Maybe this utilizes MLflow, https://mlflow.org/?
  - Experiment and plan: explore and visualize
  - Register project and track data science code required to reproduce runs on any platform
  - Save binary artifacts to compare performance between different models
  - Model registry: store annotate, discover, and manage models in a central repository
Implement DataSet for PyTorch and TensorFlow
- Clinical use-case 1: Chest X-Ray classification
- Clinical use-case 2: ????