[EPIC] Apply Federated Learning into XNAT

[dwang-usyd]

Hypothesis Statement

• Funnel Entry Date: 01/05/24
• Epic Name: Federated Learning using XNAT and NVFlare
• Epic Owner: Dongang Wang

Description

NVFlare has been proven to be useful when triggering federated learning training. However, there have been communication issues regarding the connection between the central server and clients and the data loading process when applying federated learning among different institutes.

XNAT works as a universal imaging repository, which also supports functions like dcm2niix, registration, and neural network algorithm training based on MONAI. By applying NVFlare into XNAT, both connection and data issues could be addressed.

After reviewing the process of XNAT triggering computation and the process of provision NVFlare server/client nodes, we decided to apply NVFlare as a slave image that could be handled by XNAT.

Features

• [ ] Docker image equipped by NVFlare and related neuroimaging analysis tools (FSL, MIRTK, N4, etc.)
• [ ] XNAT works as the master container, which triggers the slave container equipped by NVFlare
• [ ] Stable and safe communications between central server node and client nodes
• [ ] User interface to assist project leads to define neural networks using MONAI and upload custom models
• [ ] Directly utilise data stored on XNAT with permission
• [ ] Project participants are able to visualize the training/testing process through XNAT

[github-actions[bot]]

Features

[github-actions[bot]]

Stories