MRI Together

Global workshop on open, reproducible, and inclusive MR research

Program Book

MRITogether25 - “Breaking Boundaries”

This year marks the fifth edition of the workshop since 2021 with a focus on “Breaking boundaries”. We want to discuss how to go beyond traditional limits in every dimension of MRI. We will explore the full spectrum of imaging—from the brain to the whole body, from ultra-high to ultra-low-field—while expanding horizons geographically and professionally, engaging participants from every corner of the globe and every role in the MRI community, including radiographers. By fostering open science, championing grassroots initiatives, and advancing science communication from academia to the wider world, we aim to create a truly inclusive, innovative, and boundary-defying forum for the future of MRI. Of course, we will also highlight the latest open source tools for acquisition, reconstruction, and analysis that have been developed by the MR community.

Day 1: December 9th

MRI Together Opening Session ! (04:30 - 04:45 UTC)

Opening words by the MRI Together 2025 Organising Committee.

Open science in MRI: what is missing? (04:45 - 05:30 UTC)

Francesco Santini, University of Basel, Switzerland
Open and Reproducible Science in general, and in our specific field of MR, has come a long way in the last few years, pushed both from a shift in approach of the whole scientific field, and from the increased relevance of software and computer science, fields in which openly releasing the methods is the norm. Sessions fully dedicated to open-source software and hardware are now commonplace in our congresses, and open-source tools have become irreplaceable in every researcher’s toolbox. So, is the problem solved? Was the reproducibility crisis in MRI averted? Or are we fully in a reproducibility crisis right now, but we are too blind even to see it? Or, on the contrary, are open and reproducible science practices actively harming our field, putting ideology above scientific achievements and technological advances? In this talk, we will discuss what exists, what is still missing, and, even more relevantly, what is important, what is up to us, the single researcher, to do and what is outside of our hands. Regardless of anyone’s opinion, open and reproducible science is currently in the zeitgeist, and we are committed to make the most of it.

Deep learning techniques to enhance MRI research (06:00-06:30 UTC)

Jianxun Ren, Changping Laboratory, Beijing, China
Neuroimaging is rapidly entering the era of big data, yet existing preprocessing pipelines still require hours or even days to process a single scan. To address this bottleneck, we developed DeepPrep, an AI-empowered preprocessing pipeline for structural and functional MRI that is open-access, efficient, and robust. DeepPrep delivers a 10-fold acceleration, reducing preprocessing time to approximately 15 minutes without compromising quality. In this talk, I will introduce the development of DeepPrep and highlight its applications in clinical research and personalized neuromodulation targeting for Parkinson’s disease. Related work has been published, or is forthcoming, in Nature, Nature Methods, and Nature Neuroscience.

Guifeng Zhai, Experimental Research Center of Medical and Psychological Science, School of Psychology, Third Military Medical University, Chongqing, China

Open-minded AI deployment in MRI and Posters Presentation A (07:00-08:00 UTC)

Coordinated by Zhiyi Chen (Third Military Medical University, Chongqing, China) and Juan Pablo Meneses (Monash University, Melbourne, Victoria, Australia)

MRI Together Opening Session! (recorded) (15:00 - 15:15 UTC)

Opening words by the MRI Together 2025 Organising Committee.

Open science in MRI: what is missing? (recorded) (15:15 - 16:00 UTC)

Fairness in Cardiac MR analysis: Tackling Biases Whilst Maintaining Performance (16:30 - 17:00 UTC)

Andrew King, King’s college London, UK
AI can be used to automate the segmentation of cine cardiac MR data. However, whilst performance overall is good, a closer look reveals underperforming subgroups of patients, notably those who were underrepresented in the training data of the AI model. Bias mitigation algorithms exist to tackle these inequalities, but they often come with an assumption and expectance of a trade-off between fairness and accuracy. We show how this trade-off can be avoided by coming to a detailed understanding of the source and nature of the bias, resulting in a model that is both fair and maintains high performance for all groups.

Open Science with Sparse Data: Physics-Informed Neural Networks Applied to MRI (17:00 - 17:30 UTC)

Francisco Sahli Costabal, Pontificia Universidad Católica de Chile, Chile
Magnetic resonance imaging (MRI) can provide rich data that can enable the creation of digital twins to help diagnose and plan treatments of specific organs. However the acquisition times for limit the amount of data that can be acquired, resulting in sparse measurements. In this talk, I will show how physics-informed neural networks can enhance the information obtained from the images with biophysical knowledge. I will show applications in cardiac electrophysiology, cardiac mechanics and cardiovascular hemodynamics that demonstrate the potential of this approach.

Challenges of AI in open science and Posters Presentation B (17:30-18:00 UTC)

Coordinated by Guillermo Sahonero Alvarez, Pontificia Universidad Católica de Chile, Chile

Day 2: December 10th

Kh Tohidul Islam, Monash University, Australia
Ultra-low-field (ULF) MRI presents unique opportunities for accessible neuroimaging, but its impact depends critically on how data are collected, curated, harmonized, and shared. In this talk, I will present our experience designing the ULF-EnC MICCAI 2025 Challenge, the first initiative to release a structured, ethics-approved dataset of paired 64mT and 3T MRI scans. I will discuss practical considerations in data quality control, preprocessing, anonymisation, harmonisation, and licensing; the logistics of building a FAIR-aligned challenge; and the lessons learned in enabling reproducible evaluation while protecting participant privacy. This session will offer practical strategies for responsible data sharing in emerging MRI domains.

Guided discussion (05:00 - 05:30 UTC)

Coordinated by Stanley Norris, Monash Health, Australia.

Luis Marti-Bonmati, Hospital Universitario y Politécnico La Fe de Valencia, Spain
EUCAIM is advancing a Europe-wide federated infrastructure that enables secure sharing and analysis of cancer imaging data. Recent developments include the completion of its core federated services, the setup of local data nodes, and the release of the Federated Analysis Toolbox for privacy-preserving, distributed AI experimentation. The project has also defined clinical use cases and a benchmarking test set to support robust validation of AI tools. This infrastructure is highly relevant for MRI data sharing, as it provides governance, interoperability, and technical mechanisms suited for multi-institutional imaging datasets. As EUCAIM moves into its sustainability phase, it is preparing to become a European Digital Infrastructure Consortium (EDIC), ensuring long-term collaboration and scalable access across Europe.

Guided discussion (06:00 - 06:30 UTC)

Coordinated by Stanley Norris, Monash Health, Australia.

Agentic MR sequence development - Leveraging LLMs with MR tools and tests for physics-informed sequence development (07:00-07:30 UTC)

Moritz Zaiss University Clinic of Erlangen and Department of High Field Magnetic Resonance, Max Planck Institute for Biological Cybernetics, Tübingen, Germany
Can LLMs code valid MR sequences in PyPulseq? Yes (most of the time) under the condition: with „thinking“, and as Agent4MR with feedback and test tools it becomes more reliable. If used naively: fails massively. Students will use it anyway, so better teach them. If used rightly: Great assistance, productivity boost. An autonomous MR researcher is at least conceivable. For better human benchmark we would need a Pulseq Olympiad.

Quality control of AI methods: Evidence and Evaluation (07:30-08:00 UTC)

Gaël Varoquaux, National Institute for Research in Digital Science and Technology, France

The Tenure-Track Bottleneck: Is the Academic System Broken? Compare the Struggles from Country to Country? (08:00-09:00 UTC)

Panel discussion among Andrada Ianus (King’s College London, UK), Maria Eugenia Caligiuri (University Magna Graecia Catanzaro, Italy), and Marilou Ramos-Pamplona (University of Liege, Belgium).

Transferring Technology in a LMIC setting (14:00 - 14:30 UTC)

Andrew Webb, Leiden University, The Netherlands One of the major goals of our and many other research groups is to enable sustainable local production and maintenance of low-field MRI units in different LMICs around the world. The first step to this goal entails designing hardware and software to obtain the highest possible image quality. The next step is to make this technology widely available via open-source repositories and educational training. However, there are a large number of factors which need to be considered for this second step to be successful, which will be the topic of the talk.

MRI Advances from the Global south (14:30 - 15:00 UTC)

Joshua Ametepe, Cardiff University, UK

MRI in the LMICs (15:00 - 16:00 UTC)

Panel discussion among Cowles Chilongulo (Queen Elizabeth Central Hospital), Lydia Sekoli (University of Pretoria, South Africa), Tom Esweu (Makerere University, South Sudan), Joshua Ametepe (Cardiff University, UK).
This session will explore how innovative technologies can be successfully adapted and scaled to address critical development challenges in healthcare. How do we get great technology to work—and keep working—in Low- and Middle-Income Countries (LMICs)? It is not just about shipping the hardware but ensuring these innovations truly benefit the communities they are meant for. The panel will share real-world stories and practical solutions, from the training of local teams and handling patients to finding ways to make these projects lasting, impactful and tailored for LMIC.

Agentic MR sequence development - Leveraging LLMs with MR tools and tests for physics-informed sequence development (recorded) (16:30-17:00 UTC)

Quality control of AI methods: Evidence and Evaluation (recorded) (17:00-17:30 UTC)

Gaël Varoquaux, National Institute for Research in Digital Science and Technology, France

The Tenure-Track Bottleneck: Is the Academic System Broken? Compare the Struggles from Country to Country? (recorded) (17:30-18:30 UTC)

IMAGINE: Building MRI Capacity Through Open-Source Low-Field Training (19:00 - 19:300 UTC)

Marina Fernandez-Garcia, MRILab, Institute for Molecular Imaging and Instrumentation (i3M), Consejo Superior de Investigaciones Científicas & Universitat Politècnica de València, Valencia, Spain. And Consortium for Advancement of MRI Education and Research in Africa (CAMERA), Canada
This talk presents the IMAGINE Summer School, an ESMRMB–CAMERA initiative designed to address the global need for accessible MRI training and technology. We will discuss why initiatives like IMAGINE are necessary for strengthening MRI capacity in low-resource settings and show how two low-field, open-source MRI scanners were built in parallel by trainees in Montreal and Cape Town. The session will highlight key lessons learned—from hardware design, reproducibility, to team-based problem solving—and reflect on how hands-on training accelerates local innovation. It will conclude with our plan for expanding the program and developing sustainable pathways for accessible imaging.

Responsible AI for Brain MRI (19:30 - 20:00 UTC)

Mariana Bento, Department of Biomedical Engineering, Department of Electrical and Software Engineering, Schulich School of Engineering, University of Calgary, Canada
In this presentation, Dr. Bento will discuss key principles and challenges in developing responsible AI applications for brain MRI, highlighting the importance of open, multicenter and heterogeneous datasets. The talk will showcase applications in motion mitigation and the study of sex differences, illustrating how AI can address real clinical needs while avoiding unintended biases. Dr. Bento will also cover advances in model explainability and rigorous evaluation practices that support fairness and generalizability, contributing to trustworthy deployment in neuroimaging and healthcare.

Day 3: December 11th

AI in MRI: Pitfalls and successes (04:30 - 05:00 UTC)

Reproducible AI in MRI applications (05:00 - 05:30 UTC)

Francisco Sahli Costabal, Pontificia Universidad Católica de Chile, Chile

Randomly matching attendees to network (05:30-06:30 UTC)

Coordinated by Juan Pablo Meneses (Monash University, Melbourne, Victoria, Australia)

Science communication from outside the research world (07:00 - 07:40 UTC)

Sanam Assili, Harvard Medical School, USA
This talk explores how open-science practices and accessible communication can connect radiographers, radiologists, graduate students, technologists, engineers, and researchers across the MRI ecosystem. Drawing on Sanam Assili’s international experience across academia, industry, biotech, and high-field MRI systems at Harvard—and her work launching early ESMRMB podcast episodes and community events during COVID—the session highlights practical ways to share expertise, strengthen training, and improve transparency outside traditional research roles. Using examples from the podcast series and collaborative initiatives she helped develop, the talk illustrates how open, community-driven communication accelerates learning, expands access to knowledge, and supports MRI professionals at every stage of their career.

MRI artifacts identification game (07:40-08:20 UTC)

Cliff Mokua (Sonar Imaging Centre, Kenya), Ivy Ohuma (AIC Cure International Hospital, Kenya)
Join us for an interactive MRI Artifact Challenge Game where you will test your skills in identifying common image artefacts.We will have a fun and educational way to improve your artifact recognition abilities by going through live examples, discuss the causes, and answer your questions.

Radiographer led MRI research (08:20 - 09:00 UTC)

Christian Emery, Birmingham City University, UK
This talk explores how radiographers can lead and contribute to MRI research through accessible open-science practices. The talk shows how open data, transparent methods and collaborative networks reduce the traditional barriers that prevent radiographers and students from engaging in MRI research, especially in resource- limited settings. The session outlines practical entry points for developing research skills, joining existing projects and initiating new ones without requiring extensive funding or dedicated scanner access. The aim is to demonstrate how open science can broaden participation, strengthen research quality and support a more inclusive MRI research culture across Africa.

Cloud MR: Transitioning MRI Research From Static Infrastructure to a Fully Serverless and Scalable Cloud Native Ecosystem (14:00 - 14:40 UTC)

Eros Montin, New York University, USA
Cloud MR is a cloud native open source framework designed to modernize MRI research by replacing traditional server based architectures with a fully serverless and event driven environment. In this presentation, we describe our transition from an EC2 and Kubernetes deployment to a flexible model built on AWS Lambda, AWS Fargate, API Gateway, and Amazon S3. This architecture enables automatic scaling, eliminates idle compute costs, simplifies multi center deployments, and ensures reproducible processing of MRI data across diverse environments. We highlight two components of the platform: MR Optimum, and TESS temperature estimation. Each of these tools can run in the cloud or locally using exactly the same Python code. Through practical demonstrations, including the execution of MRI raw data in Google Colab, we show how Cloud MR enables researchers to develop, test, and deploy advanced MRI pipelines with minimal setup and complete consistency. Cloud MR aims to democratize high performance MRI computation and provide a reproducible and scalable foundation for collaborative imaging research.

BART: Sequence Programming, Simulation, Reconstruction (14:40 - 15:20 UTC)

Martin Uecker and Daniel Mackner, Graz University of Technology, Austria
In this session, we will give an introduction to the BART toolbox and show features related to sequence generation, simulation, and reconstruction. Link: https://github.com/mrirecon/bart-workshop/tree/master/mri_together_2025

Go with the Flow: Using Extensible and Shareable Motion Phantoms in KomaMRI (15:20-16:00 UTC)

Carlos Castillo-Passi, University of Stanford, USA
This tutorial will show how to model realistic motion in MRI using the new phantom framework in KomaMRI. We will introduce KomaMRI’s motion model, in which motion is built from modular actions, time curves, and spin subsets. This allows us to simulate a wide range of motion, from simple translations and rotations to complex trajectories such as cardiac motion, blood flow, or head motion. We will demonstrate how to reuse existing motion phantoms and how to create new ones for testing pulse sequences and motion-robust reconstruction methods. The session will combine a guided walkthrough of key concepts with live demos in Julia, focusing on practical workflows that participants can adapt to their own applications. Optional preparation for attendees: install Julia and the KomaMRI.jl package in advance, following the instructions at https://juliahealth.org/KomaMRI.jl/dev/how-to/1-getting-started/. Other links:

Presentation and material: https://github.com/cncastillo/KomaMRI_MRITogether2025
Documentation: https://juliahealth.org/KomaMRI.jl/dev/*

Analysis‑Grade MRI at Scale: the NiPreps experience (16:30 - 17:00 UTC)

Oscar Esteban, University of Lausanne, Switzerland
Consistent with this year’s call to “break boundaries,” my talk will depart from the limitations set by analytic variability in MRI. Even when two labs start from similar raw data, they can still reach incompatible conclusions because they followed different trajectories in the multiverse of reasonable research workflows. I will introduce NiPreps (www.nipreps.org), an ecosystem of open‑source preprocessing workflows designed to turn raw MRI into analysis‑grade data—minimally processed and safe to use directly in statistical modeling. By turning preprocessing into standardized infrastructure, we spend less time wrestling with pipelines and more time doing science to produce more comparable results.

An Open Web Tool for Pulse Sequence Design, Simulation, and Motion Exploration in Body MRI (17:00 - 17:30 UTC)

Pablo Villacorta-Aylagas, Laboratorio de Procesado de Imagen, Universidad de Valladolid, Spain
In this talk, I will present an open-source, web-based environment for interactive pulse-sequence design and simulation, tailored to explore motion effects in body MRI. Through two hands-on examples, we will examine how flow and deformation interact with different acquisition strategies: (1) flow visualisation in GE-EPI versus bSSFP exploiting time-of-flight effects, and (2) motion sensitivity in SE-EPI with unipolar versus flow-compensated diffusion gradients in a myocardium phantom. These simulations highlight how sequence design fundamentally shapes motion sensitivity and provide an intuitive platform for researchers to understand, teach, and prototype motion-robust methods. The tool aims to lower barriers to reproducible experimentation and foster a more transparent and collaborative approach to body-MRI sequence development.

Cones for 5D Free-Running bSSFP at 0.55 T (17:30-18:00 UTC)

Martin Nicoletti, Lausanne University Hospital (CHUV), University of Lausanne (UNIL), Lausanne, Switzerland
At 0.55 T, the intrinsically limited SNR increases the need for homogeneous k-space sampling in free-running 5D bSSFP. We introduce a dual-echo 3D Cones trajectory implemented in Pulseq and compare it to radial and bSTAR. Cones provides more uniform sampling and a PSF structure that is better suited for sparse reconstruction, leading to reduced artifacts and improved 5D image quality in vivo.

How to make diffusion work in the body? (18:00-18:30 UTC)

Dimitrios Karampinos, École Polytechnique Fédérale de Lausanne (EPFL)
Diffusion is a powerful contrast mechanism in modern body MRI protocols, especially will focusing on oncological questions. Physiological motion is a major challenge when applying diffusion in the body. The present presentation will introduce technical advances on how to deal with motion in body MRI with a focus on motion compensated diffusion encoding waveforms. Examples will be presented in the context of prostate, kidney and liver diffusion imaging.

Towards a Robust Automatic Quantification of Enlarged Perivascular Spaces from Brain MRI (18:30-19:00 UTC)

Roberto Duarte Coello, The University of Edinburgh, UK
Perivascular Spaces (PVS) are microscopic compartments around cerebral blood vessels, through which Cerebrospinal Fluid (CSF) can circulate. Preliminary studies have shown a possible relationship between enlarged PVS and early cognitive decline. Computational quantification remains challenging partly because these small objects’ information easily distorted by imaging artefacts. In addition, other markers of cerebrovascular dysfunction, such as White Matter Hyperintensities (WMH), often coexist, may be misclassified as, or obscure PVS.

Day 4: December 12th

Mapping the spatiotemporal continuum of structural connectivity development across the human connectome in youth (04:30 - 05:00 UTC)

Zaixu Cui, no recording, Chinese Institute for Brain Research, Beijing, China
Childhood and adolescence involve protracted remodeling of cortico-cortical structural connectivity, yet its connectome-wide developmental organization remains unclear. Using diffusion MRI across three youth cohorts, we identified a developmental continuum aligned with a sensorimotor–association (S–A) connectional axis, spanning early increases in sensorimotor connectivity to later increases in association connectivity, with a transition around age 15.5. This axis captured spatial variation in links between connectivity, cognition, and psychopathology, with psychopathological effects primarily localized to association connections. These findings provide a normative framework for understanding developmental variability in youth brain connectivity.

Short talk on the collaboration and contributions from early-career researchers (05:00 - 05:30 UTC)

Vincent Yuan (The University of New South Wales, Sydney, Australia), Zhilin Ren (Experimental Research Center of Medical and Psychological Science, School of Psychology, Third Military Medical University, Chongqing, China), Bashir Bolaji Tiamiyu (University of llorin, llorin, Nigeria), Caleb Onoja Akogwu (Wuhan Botanical Garden, Chinese Academy of Sciences, Hubei, China

Round table (05:30 - 06:30 UTC)

Live discussion among Zaixu Cui, Zhilin Ren, Caleb Onoja Akogwu, Vincent Yuan, and Bashir Bolaji Tiamiyu.

Closing session: Engineering low-field MRI and translating it into clinical practice (07:00 - 08:00 UTC)

Zhiyong Zhang School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China
Low-cost portable MRI has the potential to expand MRI access to rural areas and low and middle-income countries, provide population health monitoring, and be deployed in clinical settings such as intensive care units. However, lots of engineering effort are needed to increasing the clinical values. In the talk, I will brief introduce our experience to translating low field MRI techniques into diagnostic medicine and clinical practice, from hardware engineering, pulse sequence engineering to design engineering. Specially, experiences installing low field MRI and future challenges will be discussed.

Mapping the spatiotemporal continuum of structural connectivity development across the human connectome in youth (15:00 - 15:30 UTC)

Short talk on the collaboration and contributions from early-career researchers (recorded) (15:30 - 16:00 UTC)

Round table (16:30 - 17:00 UTC)

Another live discussion.