WP4: Delivering the AI engine behind Bigpicture’s future

Turning data into usable intelligence

Bigpicture’s ambition has always relied on a simple principle: data alone is not enough. To create value, users need tools that help them explore, validate and put that data to work. The role of WP4 has been to create the science and the technology needed to build these novel tools. Over the past four years, the team has developed the AI models, methods, and pipelines that will enable a wide range of functionalities on platform, from automated quality control and artefact detection to image retrieval and annotation assistance. All these features form the engine that will make Bigpicture usable at scale, scientifically valuable, and clinically relevant. Francesco summarises it clearly: “WP4 provides the AI ecosystem that makes the whole platform more than just a data repository.”

This work reflects the combined scientific contributions of WP4, with Raphaël Marée’s team at ULiège driving major developments in Cytomine, large-scale annotation workflows, and CBIR.

Where we are now: from development to delivery

Despite technical challenges and delays, including the unexpected bankruptcy of the Cytomine company, WP4 is now approaching one of its biggest project milestones. By January, several flagship tools will be integrated into Cytomine or available in local instances:

• Whole-slide image registration (Fraunhofer MEVIS)
• Content-based image retrieval (CBIR) (HES-SO / ULiège)
• AI-assisted cell and region annotation (University of Warwick)
• Artefact segmentation for QC (Radboudumc)

This represents the culmination of years of methodological development, coding, optimization and coordination across multiple partners. “We have delivered the methods, published the science, and built the tools. Now we’re putting them in the hands of users,” Francesco says.

A stronger Cytomine: quietly rebuilt behind the scenes

A crucial part of this milestone is the evolution of Cytomine itself. The ULiège team, led by Raphaël Marée, has taken ownership of the entire codebase, hired new engineering staff, and worked intensively to:

• scale the system to millions of slides
• integrate with Bigpicture’s data backbone
• implement authentication & authorization
• create the “App Engine” for AI deployment
• prepare an initial collection of deployable models
• support DICOM, complex annotations and multi-centre workflows

Much of this work happens out of view, but it is foundational for the tools WP4 is delivering. Francesco highlights the effort: “Liège had to rebuild critical components almost from scratch, and they did an impressive job.”

What happens next: pathologists take the lead

With WP4’s tools becoming available in Cytomine, Bigpicture enters a phase where clinical validation becomes the priority. Pathologists can now explore the tools, design validation studies, and define real use cases that matter in daily practice. This builds on previous collaborations between WP3 and WP4, such as the kidney biopsy retrieval study. As Francesco puts it: “We are moving from performing analytical validation to investigating clinical relevance.”

Looking forward: from task-specific AI to foundation models

The next major step is the start of Bigpicture’s foundation model effort, expected to kick off in early 2026 once agreements are finalized. A large-scale, European-trained model would significantly increase robustness, reduce the need for constant retraining, and support a broad range of clinical and toxicological applications. Francesco: “If we can train a model on millions of heterogeneous slides from multiple European medical centers, that becomes a cornerstone for the entire field.”

WP4 and the sustainability of Bigpicture

WP4’s outputs directly strengthen Bigpicture’s long-term value: validated tools increase the scientific utility of the platform, and indirect-access mechanisms for running algorithms offer a powerful sustainability route, especially once foundation models become available. Taken together, these are just some of the ways to help ensure the platform remains useful and relevant beyond the funded project.

Call to action: use the data and the models

As WP4 wraps up its core deliverables, the message to partners and pathologists is clear: the tools are ready, now it’s time to use them. Francesco: “My biggest hope is that pathologists will take the lead: upload data, use the tools, validate models, and drive the clinical impact. We’ve built the foundation, now we can build on it, together.”