Dr Thomas Close1,2,3, Dr Mahdieh Dashtbani-Moghari1,2,3, Mr Antoine Blachair4, Mr Aswin Narayanan2,5, Dr Oren Civier2,6, Mr David McFarlane7, Prof Fernando Calamante1,3, Dr Ryan Sullivan1,8
1Department of Biomedical Engineering, The University of Sydney, Sydney, Australia, 2National Imaging Facility, , Australia, 3Sydney Imaging Core Research Facility, The University of Sydney, Sydney, Australia, 4Neuroscience Research Australia, Sydney, Australia, 5Centre for Advanced Imaging, The University of Queensland, Brisbane, Australia, 6Swinburne Neuroimaging, Swinburne University of Technology, Melbourne, Australia, 7ResTech, The University of New South Wales, Sydney, Australia, 8Research Tech, ICT, The University of Sydney, Sydney, Australia
Introduction
The Australian Imaging Service (AIS) is a network of federated data repositories for imaging research based on the XNAT informatics platform (http://xnat.org). As part of the AIS project, a framework to deploy analysis pipelines to XNAT’s Container Service (CS) has been developed. The framework enables the AIS team to maintain a suite of well-accepted analysis methods and users to contribute their own pipelines to be run by the XNAT CS.
Methods
The core logic for the AIS pipelines framework has been developed as part of Arcana (https://arcana.readthedocs.io) and uses Arcana for XNAT I/O and format conversions. Dockerized pipelines for the XNAT CS are built from recipes specified in YAML. The recipes for core and community developed pipelines are stored on GitHub in the repositories http://github.com/Australian-Imaging-Service/pipelines and https://github.com/Australian-Imaging-Service/pipelines-community, respectively. GitHub actions are used to trigger the Arcana build and push the Docker pipelines to the GitHub container registry from which they can be pulled to the AIS nodes.
Results
A suite of community-developed pipelines for neuro MRI analysis were implemented within the framework. XNAT CS Docker images for arbitrary shell commands, Pydra workflows (https://pydra.readthedocs.io) and Brain Imaging Data Structure (BIDS) apps (https://bids.neuroimaging.io) can be generated directly from YAML recipes. Provenance is stored alongside generated derivatives that details the location of the source data and all processing steps used, from which exact analysis can be reproduced.
Conclusion
The proposed framework enables state-of-the-art analysis methods to be provided to AIS users in a convenient, secure and reproducible manner.
Biography:
Dr Close is the National Imaging Facility (NIF) Informatics Fellow at the University of Sydney, where he leads the Analysis Pipelines Stream of the Australian Imaging Service (AIS). In this role, he oversees the development of an integrated library of biomedical imaging analysis workflows, and a framework to execute the workflows over large and dispersed datasets using cloud and high-performance computing clusters.