‘I started making 3D reconstructions of chicken hearts back in 2006 during my second year of study, right in the anatomy department's dissection room at AMC,’ De Bakker explains. ‘After my doctorate, I was offered a position to create a three-dimensional atlas of human embryos, under the guidance of Professor Antoon Moorman. That became the foundation of everything I do now.’ 

De Bakker spent over five years on the so-called 3D Embryo Atlas – a digital model that allows you to follow the complete embryonic development up to 10 weeks of pregnancy. ‘You can view the embryo on your screen, rotate it, and click on and toggle all organs on and off. For the first time, that early human development was depicted in such an accessible way.’ 

In this project, she also collaborates closely with international partners within initiatives like the Human Organ Atlas Hub. ‘There, you see incredibly smart people at the table – postdocs, mathematicians, physicists – who are eager to build, but they hit the same wall: there's not enough thorough, validated anatomical data to train AI on. And that's exactly what we're focusing on.’ 

What does your project involve? 

‘In this project, we're working on automatically segmenting the vast amount of 3D image data we've collected over the years. It truly involves petabytes of data. It's impossible for a human to manually sift through all of that. That's where we need AI,’ explains Bernadette. 

But artificial intelligence needs examples to learn from – the so-called 'ground truth data'. ‘An AI doesn't inherently know what a blood vessel or a nerve is. You have to teach it first. And for that, you need specialists who can recognize and mark those structures. But doctors often don't have the time.’ 

The solution De Bakker and her team are exploring involves 'citizen science': engaging well-trained citizens – such as medical students or volunteers with a biomedical background – to assist in annotating the data. ‘This way, we're building a reliable dataset step by step, which not only benefits us but also others moving forward.’ 

Why did you apply for the Impact Call? 

‘For initiatives like ours, it's crucial to have the opportunity to invest in translating science into broad applicability,’ says De Bakker. ‘Thanks to the Impact Call, we can now make real progress in developing a workable AI segmentation model and establish a sustainable infrastructure for data analysis. We can train people to assist in labeling the data and provide technical support like laptops to further develop the models.’ De Bakker emphasizes that she doesn't carry the project alone. ‘For instance, I co-wrote the application with PhD candidate Puck Mulder, who made a significant contribution. Such collaboration is essential. You can't do it alone.’ 

What do you hope for your project? 

‘I hope this project marks the beginning of something bigger,’ De Bakker asserts. ‘We're building a fundamental infrastructure for biomedical research into human development. And that foundation is essential for various other innovations – in AI, medical education, diagnostics.’ 

She already sees significant interest: ‘Within international consortia, I see a lot of enthusiasm, but also frustration because they can't progress without reliable data. Our work can help fill that gap. If we set this up well, it can become a standard that the whole world benefits from.’ 

Additionally, De Bakker sees opportunities to make the project accessible to students and doctors in training. ‘It's not just research; it's also education, collaboration, technology, and clinical practice. Those connections are rare and valuable.’ 

Finally, she hopes for sustained attention. ‘I want this project to be more than just a one-off. I want it to become a model for how the biomedical community can embrace new technologies. We shouldn't just use AI; we should understand it – because we laid the groundwork ourselves.’