This year’s ISMRM felt different in several ways.
Not only because it was the first time the congress was held in Cape Town — and in Africa — but also because many of the conversations happening across the event seemed to move beyond image quality and model performance alone, and increasingly toward the operational realities behind advanced imaging research.
Walking through the congress floor, it was impossible to ignore the scale of innovation happening across MRI hardware, reconstruction methods, AI-assisted analysis, and advanced medical imaging workflows. The major MRI manufacturers showcased increasingly sophisticated systems and emphasized not just image quality, but also flexibility: the ability for researchers to customize sequences, reconstruction methods, and post-processing workflows more easily than before.
At the same time, a different type of conversation kept surfacing repeatedly in discussions with researchers and imaging teams.
How do we actually scale all of this?
One of the clearest recurring themes throughout the week was the growing difficulty of sharing imaging data across institutions, teams, and environments.
While interoperability has been a long-standing challenge in medical imaging, many conversations highlighted how fragmented imaging research workflows still are when different institutions, systems, vendors, and infrastructures need to work together.
Interestingly, the issue was often less about image analysis itself and more about the operational friction surrounding the data.
Who can access it?
Where is it stored?
How can it be shared securely?
How do different systems interact with one another?
Those operational questions are becoming increasingly important as multi-site imaging studies and AI-driven medical imaging workflows continue growing in complexity and scale.
Several discussions also reflected a broader concern around infrastructure and data sovereignty, particularly among European researchers and institutions. There was noticeable hesitation around relying entirely on large US-based cloud providers for storing and managing sensitive medical imaging data.
In many cases, this was not framed as a purely technical issue, but as a strategic and regulatory concern around long-term control, governance, and infrastructure independence.
That context also helps explain why federated learning appeared in so many conversations throughout the event.
As imaging data sharing restrictions become more complex, especially across countries and institutions, federated approaches are increasingly viewed as a practical way to collaborate without requiring centralized access to sensitive datasets.
While federated learning in medical imaging has been discussed in research circles for years, it felt notably more present this year as organizations continue searching for ways to scale AI research without moving large amounts of imaging data across environments.
The growing interest around federated learning also reflects a broader industry challenge: balancing collaboration, scalability, compliance, and governance at the same time.
Another recurring point was the growing recognition that current approaches to imaging research infrastructure may simply not scale much longer.
Many researchers openly acknowledged that continuously expanding local storage and compute infrastructure inside universities is becoming increasingly difficult to sustain over time.
As MRI datasets grow larger and AI imaging workflows become more computationally intensive, the operational burden of maintaining on-premise infrastructure also grows.
At the same time, moving toward cloud-based medical imaging infrastructure introduces new concerns around governance, permissions, interoperability, and trust.
In many ways, the industry appears to be reaching an interesting transition point.
The scientific capabilities around MRI, imaging biomarkers, and AI-assisted analysis continue advancing rapidly. But operational infrastructure, governance models, and scalable imaging workflows are still catching up in many environments.
That gap between scientific innovation and operational execution was one of the most interesting undercurrents throughout ISMRM 2026.
One interesting observation throughout the congress was that many conversations were less focused on clinical trial workflows specifically and more centered around research flexibility and infrastructure enablement.
This also became visible in how MRI manufacturers positioned themselves during the event.
Beyond showcasing new hardware and image quality improvements, several vendors increasingly emphasized how easy their systems were to customize for research purposes — including programmable sequences, reconstruction methods, and post-processing capabilities.
That shift reflects a broader reality across advanced imaging research: innovation is no longer only about acquiring better images. It is increasingly about creating environments where research teams can operationalize, scale, share, and reproduce complex imaging workflows more efficiently.
For organizations working in imaging infrastructure and imaging data management, these conversations matter because they reflect how the priorities of imaging teams are evolving.
Beyond algorithms and visualizations, there is growing awareness that research scalability increasingly depends on mature operational systems capable of supporting collaboration, governance, interoperability, permissions management, and long-term imaging data management.
These are precisely the types of operational challenges platforms like QMENTA are increasingly designed to support across imaging research and clinical environments.
Events like ISMRM remain incredibly valuable precisely because they bring these conversations together in one place — from MRI physicists and AI researchers to infrastructure providers, imaging teams, and clinical research organizations.
And while the technologies showcased this year were impressive, some of the most important discussions were ultimately about something less visible: the operational infrastructure required to make advanced imaging research scalable in the real world.
Federated learning is an AI training approach where models are trained across multiple institutions or datasets without requiring sensitive medical imaging data to be centrally shared or transferred.
Medical imaging data sharing is often limited by interoperability issues, regulatory restrictions, infrastructure differences, governance concerns, and data privacy requirements.
As imaging datasets, AI workflows, and multi-site collaborations continue growing, research teams increasingly require scalable infrastructure capable of supporting storage, processing, permissions management, and collaboration efficiently.
Common challenges include interoperability between systems, data sharing restrictions, infrastructure scalability, workflow harmonization, governance, and managing increasingly large imaging datasets.