Steffen Bollmann1, Kurt Sterzl3, Aswin Narayanan2
1The University of Queensland, St Lucia, Australia, 2Australian National Imaging Facility, St Lucia, Australia, 3Amazon Web Services, Inc, Brisbane, Australia
Biography:
After obtaining a Master degree in Biomedical Engineering at the Ilmenau University of Technology, Steffen completed a PhD on multimodal imaging at the University Children’s Hospital and ETH Zurich, Switzerland. Dr Bollmann then joined the Centre for Advanced Imaging at the University of Queensland as a National Imaging Facility Fellow, where he pioneered the application of deep learning methods for quantitative susceptibility mapping, in the group of Prof Markus Barth. In 2019 Steffen joined the Siemens Healthineers collaborations team at the MGH Martinos Center in Boston during a 1 year industry exchange where he worked on the translation of deep learning reconstruction techniques into clinical applications. Since joining the School of Information Technology and Electrical Engineering at the University of Queensland in 2020 Dr Bollmann develops computational methods to process quantitative magnetic resonance imaging data. As part of the ARDC platform grant AEDAPT, Steffen is developing the NeuroDesk platform – a flexible, scalable, and browser-based data analysis environment for reproducible neuroimaging.
Abstract:
Introduction:
Researchers need a wide range of tools to analyse data and address their research questions. However, training researchers to use these tools is challenging. Many of these tools require Linux and are difficult to install due to complex dependencies. Additionally, neuroimaging analysis demands significant computational power and storage. This often requires extended pre-workshop setup instructions and forces workshops to use simplified datasets because of the limitations of participants' computers.
Methods:
To overcome these challenges, we developed a flexible, cloud-based deployment of Neurodesk (www.neurodesk.org). This setup allows quick and efficient provisioning of a workshop on the cloud. We built on the Zero to JupyterHub project, which uses Kubernetes to deploy JupyterHub, and incorporated CVMFS to handle Neurodesk’s software distribution. For smaller workshops, we created a K3s-based deployment that only requires a single virtual machine, avoiding the need for a full Kubernetes setup.
Results:
This setup has been used in over 10 workshops and university courses, with overwhelmingly positive feedback from instructors. Many stated that the cloud setup made their courses more interactive and enabled the use of realistic data analysis examples. Cost efficiency was another major benefit, as resources could be scaled down when not in use, which was particularly important for GPU-intensive deep learning workshops.
Conclusion:
The Neurodesk cloud-based platform has proven to be a valuable tool for teaching neuroimaging data analysis. Based on the success of these workshops, we are currently expanding the platform to support other scientific domains such as genomics, microscopy, and astronomy.