Improving MRI Reproducibility: A Practical Guide for Researchers and Imaging Centers

Even seemingly stable measurements, such as brain volume, can vary between scanners or acquisition setups. This highlights why structured frameworks, consistent validation, and reproducible tools are so important for reliable research.

At NeuRA Imaging, in collaboration with the National Imaging Facility (NIF), we are helping build a national framework that strengthens every step of the MRI research lifecycle.

Why Standardisation Matters: Starting with BIDS

The Brain Imaging Data Structure (BIDS) has become one of the most important tools for improving reproducibility. It provides clear rules for how MRI data should be organised, named, and documented. Consistent organisation makes it easier to share data, compare results, and reproduce analyses across teams.

BIDS also preserves essential metadata – acquisition parameters, participant information, and task details – which are often lost in traditional storage systems. By adopting BIDS, researchers increase the longevity, compatibility, and reusability of their data.

What Needs to Happen Before Any MRI Study

Several prerequisites should be in place before a research MRI scan is performed, ensuring that data remains trustworthy and comparable:

• A DOI assigned to each unique instrument is a preferable practice, including hardware details, software versions, and acquisition dates
• Optimised imaging protocols validated using MRI phantoms
• Validation not only with phantoms but also in real human participants
• A clear QA and QC schedule to monitor scanner performance over time

These foundations reduce variability, making it possible to compare data across sessions, studies, and sites.

National Infrastructure: MRNet and ABIRD

Across Australia, NIF’s MRNet project is standardising MRI workflows across the national 3T scanner network. The focus includes high-quality acquisition, secure and standardised storage, and reproducible measurements regardless of where a scan takes place.

This infrastructure supports ABIRD, the Australian Biomedical Imaging Research Database, which is building a high-quality resource of healthy MRI scans. ABIRD aims to help researchers conduct large-scale studies and access reliable control datasets for future research.

Storing MRI Data: Open Source or Commercial?

MRI data is large, complex, and sensitive, so choosing the right storage platform is essential.

• Open‑source platforms (e.g., XNAT) offer flexibility and no licensing fees but require technical support and local infrastructure.
• Commercial platforms (e.g., Flywheel) simplify setup, user management, and security but involve ongoing financial commitments.

All systems must comply with ethics requirements, participant de‑identification standards, and secure‑access protocols.

QA/QC and the Importance of MRI Phantoms

Phantoms – objects with known properties – are essential tools for monitoring scanner performance. They help researchers:

• Detect scanner drift
• Identify differences between sites
• Validate performance after hardware or software changes

NIF acquired phantoms for harmonisation at each site and across the network from Gold Standard Phantoms to ensure the quality and reproducibility of images obtained on our Philips 3T Ingenia MRI scanner.

Reproducible Data Analysis with Neurodesk

Even with excellent acquisition and storage, analysis tools can introduce variability. Software versions, dependencies, and operating system differences can all affect results.

Neurodesk addresses this challenge through a fully containerised software ecosystem designed for reproducible neuroimaging. Key features include:

• A consistent suite of neuroimaging tools with all dependencies included
• A virtual desktop and JupyterLab environment accessible through the browser
• Compatibility with Windows, macOS, Linux, HPC environments, and cloud systems
• Support for FAIR‑aligned workflows and tools like OpenNeuro
• A community‑driven development model for transparency and sustainability

Why All This Matters

Together, these practices — BIDS organisation, validated imaging protocols, rigorous QC, reproducible analysis environments, and secure, standard‑aligned storage — help ensure that MRI research produces reliable, meaningful, and reusable results.

This approach enables:

• Confidence that measurements reflect real biological differences
• Accurate multi‑site clinical trials
• Long‑term preservation of data value
• Avoiding the common pitfalls that make older MRI datasets unusable today

Reproducibility isn’t a technical afterthought. It is central to high-quality imaging science.

This work was submitted as an abstract and presented as a scientific poster at the International Research Integrity Conference, held at the University of Sydney in November 2025, with the following presenting authors: Michael Green, Steffen Bollmann, Bryan Paton, and Arkiev D’Souza. You can download the poster here.