README.md

Back to Projects List

Data and model exchange across different sources

Kaapana tutorial for the 38th NA-MIC project week:

https://drive.google.com/file/d/1A7-8Ru0uTJHFFa17rZtkBpvNhJao_F7x/view?usp=share_link

Key Investigators

• Benjamin Hamm (German Cancer Research Center, Germany)
• Ünal Akünal (German Cancer Research Center, Germany)
• Markus Bujotzek (German Cancer Research Center, Germany)
• Klaus Kades (German Cancer Research Center, Germany)
• Andrey Fedorov (Brigham and Women's Hospital, USA)

Project Description

Implementations and discussion about a standardized data and model exchange between different platforms such as Kaapana and MONAI. Working on integrating Kaapana with other toolkits.

• Motivation: Running Kaapana platforms in multiple (inter-)national projects: RACOON, DART, ...
• Goal: Standarized and Federated Data Analysis / Federated Learning require standardized model exchange formats

Objective

Support standardized data and AI model I/O interfaces in Kaapana.

1. Support of various AI model sources

• Integration of MONAI Model Zoo into Kaapana
  • inference pipeline as a Kaapana workflow / as a Kaapana extension
  • training pipeline
  • generic support of MONAI Bundles (MONAI Label / MONAI Deploy / MONAI FL)
• Standardized remote model execution, execution of models from modelhub.ai within Kaapana

2. Integration/support of data sources:

• TCIA download/(upload) into Kaapana
• Integration with IDC: download of data via Google Cloud SDK

3. Integration of new analysis tools into Kaapana
4. Javascript/Python library client to communicate with Kaapana

Relate to:

• Kaapana overall

Approach and Plan

1. Support of various AI model sources

• Integration of MONAI Model Zoo into Kaapana
  • inference pipeline as a Kaapana workflow / as a Kaapana extension
  • training pipeline
  • generic support of MONAI Bundles (MONAI Label / MONAI Deploy / MONAI FL)
• Standardized remote model execution, execution of models from modelhub.ai within Kaapana
• Current progress:

2. Integration/support of data sources:

• TCIA download/(upload) into Kaapana
  • Kaapana workflow to download specific TCIA datasets
  • select to-be-downloaded dataset via UI
  • send downloaded dataset to Kaapana's PACS

Progress and Next Steps

1. Support of various AI model sources

• Integration of MONAI Model Zoo into Kaapana
  • Proof of concept: Intgration of MONAI Model Zoos spleen CT segmentation works
  • tbd: Finalize integration in Kaapana
  • tbd: Add more monai bundles
• Support of MHub
  • Completed the implementation of a workflow in Kaapana for modelhub.ai
  • Supports each model already available in mhub
  • A wrapper around the dockerfile of models in mhub
  • Ability to visualize the segmentations using Slicer, MITK or OHIF on a web browser

2. Integration/support of data sources:

• TCIA download/(upload) into Kaapana
  • Implemented service-tcia-download. Now it is possible to drag and drop a .tcia manifest file into Kaapana (in minio). This will start a workflow which downloads the data from TCIA via their REST-API. Number of workers can be set in the operator.

Illustrations

tbd

Background and References

• https://www.kaapana.ai/
• http://app.modelhub.ai/
• https://www.cancerimagingarchive.net/
• https://monai.io/