The man with 100,000 “brains”: AI’s rich gifts to science

Researchers at King’s College London are using the NVIDIA Cambridge-1 supercomputer and MONAI to create a treasure trove of open-source synthetic brain images, accelerating the use of AI in healthcare.

Jorge Cardoso may have a lot of titles, but he really lives up to his reputation because he has a lot of brains. A hundred thousand, to be exact.

Cardoso is a teacher, CTO, entrepreneur, founding member of the MONAI open source alliance, and researcher in the field of medical imaging AI. In the last role, Cardoso and his team discovered how to use AI to create high-resolution realistic 3D images of the human brain.

The King's College London researcher and CTO of the London AI Centre has made 100,000 synthetic brain images freely available to healthcare researchers - a treasure trove that could accelerate our understanding of dementia, aging or a variety of brain diseases.

Accelerating the application of AI in healthcare

“Many researchers have avoided working in the health care space in the past because they couldn’t get enough good data, but now they can,” Cardoso said.

"We hope to introduce AI research into the field of healthcare," he said.

This is a significant donation compared to the world's largest free repository of brain images: UK Biobank, which currently houses multiple brain images from more than 50,000 participants, at an estimated cost of $150 million.

Synthetic Data for Science

The images represent an emerging branch of synthetic data in healthcare, which is already widely used in computer vision for consumer and commercial applications, where open datasets containing millions of real images are already available.

In contrast, real images are scarce in the medical field. To protect patient privacy, medical images are often only available to researchers associated with large hospitals. Even then, these images often reflect only the populations served by the hospitals, not the wider population.

The important feature of this new AI approach is that it can make images on demand. Female brains, male brains, old brains, young brains, sick or healthy, etc. Just plug in what you need and the system will create it.

Although these images are simulated, they are very useful because they retain key biological features so they look and function very similar to a real brain.

Scaling up with MONAI on Cambridge-1

The work requires supercomputers that can run super software.

NVIDIA Cambridge-1, a supercomputer dedicated to groundbreaking AI research in healthcare, serves as the engine. MONAI, an AI framework for medical imaging, serves as the software fuel.

Together, they create an AI factory for synthetic data, allowing researchers to run hundreds of experiments, select the best AI model, and run inference to generate images.

“Without Cambridge-1 and MONAI, we could not have done this work, none of this would have happened,” Cardoso said.

Up to 10 times faster for massive images

NVIDIA DGX SuperPOD Cambridge-1 is equipped with 640 NVIDIA A100 Tensor Core GPUs, each with enough video memory to process one or two of the team’s 16 million 3D pixels in its sea of ​​images.

MONAI’s base modules include domain-specific data loaders, metrics, GPU-accelerated transformations, and an optimized workflow engine. Cardoso said the software’s smart caching and multi-node scaling can speed up jobs by up to 10 times.

He also praised cuDNN and "the entire NVIDIA AI software stack that makes us more productive."

More than just the brain

Cardoso is working with the UK Institute for Health Data, a national repository that hosts 100,000 brain images, and the AI ​​model will be made available to researchers to create the images they need.

Furthermore, the team is exploring how these models can generate 3D images of any part of the human body in medical imaging modalities (MRI, CAT or PET scans, etc.).

“In fact, this technique can be applied to any volumetric image,” he said, noting that users may need to optimize the model for different types of images.

Unlimited prospects

When it comes to the application prospects of synthetic image technology, Cardoso temporarily pulls himself away from his complicated thoughts and begins to introduce it enthusiastically.

The synthetic images will help researchers understand how the disease progresses over time. Meanwhile, Cardoso's team is still exploring how to apply this work to parts of the body other than the brain, and what type of synthetic images (MRI, CAT, PET) will be more useful.

The possibilities are exciting, and, as with many of Cardoso’s roles, “it can be a little overwhelming,” he said. “There are so many different things we can start thinking about right now.”