Back in 2018, Pierre Moulin – then project leader of Pathology at Novartis – looked ahead to a future of AI-assisted pathology with a crushing realization: the quality of AI applications reflects the quality of the data used to train them. And, at that time, data sets were limited to whatever each institution or company had on file.
But what if the pathology world was to pool its resources? What if industry, academics, and clinical labs could all contribute high quality data sets to a central repository, providing a library of millions of digital pathology slides and notes for AI research and development?
It might just work… but it would require a consortium of stakeholders, and funding, to get it off the ground.
Katrien Grünberg, Professor of Pathology at Radboud University Medical Center in the Netherlands, and a leader of Bigpicture, picks up the story.
How did Bigpicture come about?
Pierre Moulin initiated a call through the Innovative Medicines Initiative (IMI) – a public–private partnership organized by the European Federation of Pharmaceutical Industries and Associations. It combines funding from the European Union with private or in-kind contributions from pharmaceutical companies.
At Radboudumc we were already engaged in AI research and could clearly see that data availability was the primary bottleneck. And our scientific network included numerous colleagues who were encountering the same hurdle. So we teamed up with these colleagues and submitted a proposal. It was a competitive process, but IMI decided to fund our initiative.
There were 45 partners involved initially: ten pharmaceutical companies, various academic institutions, small and medium-sized enterprises, and professionals from a wide range of disciplines. Each component required specialized expertise, and we built a consortium that reflected that diversity.
What was the vision?
The goal wasn’t just to build a data repository, but to create a comprehensive platform. To be useful, the data needed to live within a professional ecosystem. That meant involving pathologists to ensure the data was high-quality and relevant, IT experts to build the technical infrastructure and security protocols, and AI researchers to advise on model training and data usability.
We also needed a robust software layer – an interface that could connect the hardware and the scientific users. It needed to manage user access, verify credentials, and enforce proper authorization. Ultimately, the platform is a synthesis of software, hardware, data, and – critically – a community.
How are patient privacy and data confidentiality issues being addressed?
Compliance with privacy regulations is essential. That’s why one of our work packages focuses specifically on legal and ethical frameworks. From the institutions contributing data to the researchers using it downstream, every step has to be underpinned by documentation: consortium agreement, data sharing policies, terms of use. It’s a lot of paperwork, but it serves a purpose. These agreements spell out what the data can and cannot be used for, how patient privacy is safeguarded in the case of clinical datasets, and the terms under which license agreements are granted.
Once we finalized the technical pipeline and all required paperwork – particularly the data sharing agreements – contributors could begin uploading their slides.
How do you source and select the images?
One source is curated collections – what you might think of as a pathologist’s personal archive. These are often carefully assembled, highly valuable datasets featuring rare cases or particularly illustrative examples, with thorough documentation and clinical context. We absolutely want to preserve and incorporate those collections into Bigpicture.
But we also source routine collections. If you take a typical day in the life of a pathologist, you see a cross-section of routine pathology – common conditions, negative margins, and normal lymph nodes. While these may not be as striking as the rare cases, they’re equally important for developing AI models. We need both ends of the spectrum: the exceptional and the everyday.
The high-quality data that we require doesn’t just include technically flawless slides – it means data that’s representative of real-world pathology. That includes specimens from around the globe, not just those generated in top-tier labs in Western Europe. To coordinate the inflow of such data, we’ve established a system of expert “nodes,” each focused on a specific pathology area: liver, skin, lung, renal, oncology, clinical trials, and preclinical.
Each node acts as a liaison between the central repository and data-contributing institutes. They also help ensure that the repository doesn't become dominated by any single disease area – like breast cancer – but remains balanced and inclusive.
Who are the contributors?
From the outset of the project, we’ve had a list of interested organizations who weren’t part of the original consortium but have valuable collections they’d like to contribute. We call them Slide-Contributing Third Parties (SCTPs).
Beyond the SCTPs, there’s an even wider “outer orbit” of potential contributors: other institutions that may approach us at conferences or be contacted by us directly. Once the platform is fully operational, we’ll be ready to engage with them as well.
In short, Bigpicture welcomes contributions from anyone – so long as the material is relevant, responsibly sourced, and accompanied by the necessary agreements.
Who are Bigpicture’s users, and how are they granted access?
Our primary users will fall into two main categories: scientists and developers.
When we talk about “scientists,” we mean anyone engaged in rigorous, hypothesis-driven work – regardless of whether they’re in a university or corporate setting.
AI development is often associated with commercial activity – especially when clinical data is involved – but it’s also essential for translating research into real-world impact. To truly benefit patients, those insights must be transformed into tangible products: biomarkers, diagnostic tools, new therapies. That kind of development typically happens in industry, which naturally needs to sustain itself financially to continue innovating.
That said, access to Bigpicture data isn’t automatic. Interested users must submit an application outlining their intended use of the data. We evaluate whether their proposed project aligns with Big Picture’s mission and ethical framework. If a specific dataset is requested, we also ensure that the planned use complies with the terms of our data sharing agreement.
Data access decisions may involve different review bodies. Sometimes this falls within Bigpicture’s own Data Access Committee, but in other cases – such as when data is contributed by external institutions like Radboudumc – the decision is made by the contributor’s own access committee. They assess whether the proposed use aligns with what they deem permissible for that dataset.
Once an application is approved, the user is granted a license to use the requested data under defined terms.
What does “success” look like for Bigpicture?
There is one very concrete aim for this project – to collect 3 million digital slides. The plan is to have 2 million non-clinical slides contributed by pharmaceutical companies and 1 million slides from our clinical partners.
Once we have achieved that, success will be about sustaining the platform so that it is available and relevant for future users. To achieve that, we need to build a system that allows other partners to get on board and contribute, or become a user, in the future beyond the funded action of the project.
That’s important because we know that AI development in the future will require vast amounts of relevant data of high quality and wide diversity. And needs may change over time, so the system needs to be dynamic. We might need to upgrade the platform and add more data in the future.
Ultimately, Bigpicture will succeed when a community of the brightest minds in pathology AI research has access to a reservoir of excellent data to facilitate that research. That means that access won’t be limited to large companies and institutions with lots of money.
How will you ensure sustainability of the platform?
Bigpicture will be operated as a funded, non-profit platform. At present, users can access the service free of charge. In the future, if the funding is used up, they might be charged. We need to maintain the platform and software, store the data, and reimburse the team – it’s expensive.
How will Bigpicture’s environmental impact be mitigated?
If you look at our partners, many are using renewable energy – particularly in Northern Europe. CSC in Finland provides our data management and computing ecosystem, which is powered entirely by renewable sources. That was a conscious decision.
Obviously, when users start projects using Bigpicture data, they will be using a great deal of computing power, so there will be an impact. But we are talking about scientific projects that will serve an important societal purpose, not cat videos.
How can readers of The Pathologist get involved?
We believe every pathologist should be aware of Bigpicture – not just as a passive observer, but as a potential contributor, user, and beneficiary. Whether you're interested in conducting research within your own lab, applying AI to your diagnostic practice, or using data-derived resources for educational purposes, there are multiple avenues for meaningful engagement.
One of our key partners in the project is the European Society of Pathology (ESP). Their participation ensures we remain closely connected to the pathology community, which is central to the platform’s success. We encourage all pathologists to connect with us – especially through the ESP.
Additionally, you can sign up for our newsletter, attend our webinars, or meet us in person at ESP events. There are many ways to get involved, and we’re always eager to welcome new voices into the community.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
