Driving progress in Bigpicture's data upload phase

We spoke to Julie Boisclair, EFPIA Leader of Work Package 1, and Head of Digital and Computational Pathology at Novartis, about the road so far and the lessons worth sharing. Since stepping into her role in 2021, Julie has been responsible for aligning the needs and contributions of the industry partners with the broader goals of the project. Her background in both scientific and operational domains helps bridge gaps between disciplines, and between public and private stakeholders. “Bigpicture is complex,” she says, “and one of the biggest challenges is getting everyone to speak the same language. That applies just as much to project coordination as it does to data.”

Uploading takes a team, and time

“You can’t do this alone,” Julie says as she is asked to name key takeaway for partners contributing data to Bigpicture. At Novartis, she is working closely with for instance Senior Data Manager Johann Mueller who is involved in the technical part of the data uploads. “This is not just for pathologists. You need technical staff, data scientists, IT support, legal teams, and time to get them all on the same page.”

Novartis invested early in building those connections internally, even though aligning priorities took effort. “We had to explain the value of this project to teams who don’t usually work with us. That took patience and repetition.”

Getting started: slow, messy, and necessary

While Novartis has been scanning slides for years, it didn’t mean everything was plug-and-play. “The data wasn’t as ready as we thought,” Julie explains. “We had to match labels, histo sheets, metadata, and automate as much as possible.” 

To overcome this, Novartis developed an in-house OCR (Optical Character Recognition) pipeline, now being refined with the help of large language models and AI specialists within the company. These tools are steadily reducing the burden of manual curation, transforming what used to take hours into streamlined, semi-automated workflows.

Novartis’ first uploads weren’t flawless. One study triggered even 400 errors. But with every attempt, things improved. “That’s why we started early; to make the mistakes now, not later.”

Why others are waiting. And why that’s risky.

Many partners are hesitant to start uploading, waiting for smoother tools and full automation. Julie understands, but cautions against it. “The first steps are always the hardest,” she says. “But the process gets easier the more you do it. And we’ve already ironed out many of the issues.”

Novartis deliberately tested a range of upload scenarios, from small studies to large batches, to identify weak spots. This benefits everyone. “We’re sharing what we learn, so others don’t have to go through the same struggles.”

What it takes to prepare your upload

Uploading data starts long before the actual uploading-part. “Start talking to your IT and legal departments early,” advises Julie. And don’t wait until everything is perfect, but just begin.” 

Data upload starts with tasks like:

• Checking metadata formats and slide labelling
• Making sure IP info is removed
• Setting up secure upload stations (behind firewalls, with IT approval)
• Aligning terminology with the SEND standard
• Getting internal approval from legal and data protection teams

It’s worth it (also beyond Bigpicture)

Despite the effort, Julie says the benefits go far beyond fulfilling project obligations. “This project also pushes us to clean up and structure our data. It’s helping us make our datasets AI-ready, and that has value well beyond Bigpicture.” She also points to the collaboration that’s grown from the work. “The project includes a multidisciplinairy collaboration, but it also promoted a multidisciplinary collaboration within Novartis. People in other departments got excited by the challenge. We had AI specialists help build tools on the side, and now that’s being shared across the company.”

What’s next: scaling up

Novartis is currently scaling up to upload around 400.000 slides by the end of the project. Three workstations will run uploads in parallel. Upload speed is expected to improve further once automation is in place (around September 2025). Julie sees this as the final push: “We’re in the last part of the marathon. It’s time to deliver, for the project, for the future of pathology, and for the value this platform will offer to all of us.”