KCL uses AI to generate 100,000 brain images

Jorge Cardoso wears many hats, and that’s fitting given he’s got so much brains. A hundred thousand of them to be exact.

Cardoso is a teacher, CTO, entrepreneur, founding member of the MONAI open source consortium and researcher in AI for medical imaging. In the latter role, Cardoso and his team have just discovered ways to create realistic, high-resolution 3D images of human brains with AI.

The King’s College London researcher and the CTO of the London AI Center are making 100,000 synthetic brain images freely available to health researchers. It’s a treasure that could speed up understanding of dementia, aging, or any kind of brain disease.

Accelerating AI in Healthcare

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

“We want to direct the energy of AI research into healthcare,” he said.

This is a major donation compared to the world’s largest repository of freely available brain images. The UK biobank currently holds several brain images from more than 50,000 participants, organized at an estimated cost of $150 million.

Synthetic data for science

Images represent an emerging branch of synthetic data in healthcare, something that is already widely used in computer vision for consumer and professional applications. Ironically, these fields also have access to open datasets with millions of real-world images.

In contrast, medical images are relatively rare and are generally only accessible to researchers linked to large hospitals, given the need to protect patient privacy. Even then, medical images tend to reflect the demographics served by the hospital, not necessarily the general population.

A happy feature of the new AI approach is that it can create images on command. Female brains, male brains, old people, young people, brains with or without disease. Plug in what you need, and it creates them.

Although simulated, the images are very useful because they preserve key biological characteristics, so they look and act like real brains would.

Scaling with MONAI on Cambridge-1

The job required a supercomputer running super software.

NVIDIA Cambridge-1, a supercomputer dedicated to groundbreaking AI research in healthcare, was the driving force. MONAI, an AI framework for medical imaging, provided the software fuel.

Together, they created an AI factory for synthetic data that allows researchers to perform hundreds of experiments, choose the best AI models, and perform inferences to generate images.

“We couldn’t have done this job without Cambridge-1 and MONAI, it just wouldn’t have happened,” Cardoso said.

Massive images, up to 10 times faster

An NVIDIA DGX SuperPOD, Cambridge-1 packs 640 NVIDIA A100 Tensor Core GPUs, each with enough memory to process one or two of the team’s massive 16 million pixel 3D images.

MONAI’s building blocks include domain-specific data loaders, metrics, GPU-accelerated transformations, and an optimized workflow engine. The software’s smart caching and multi-node scaling can speed up tasks up to 10 times, Cardoso said.

He also credited cuDNN and “the whole NVIDIA AI software stack which helped us work much faster.”

beyond the brain

Cardoso is working with Health Data Research UK, a national repository, to house the 100,000 brain images. AI models will also be available, so researchers can create all the images they need.

There is more. The team is investigating how the models can create 3D images of any part of the human anatomy in any medical imaging mode – MRI, CAT or PET, etc.

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

Many directions ahead

The work points in many directions which Cardoso enthusiastically described as unloading content from multiple minds.

Synthetic images will help researchers see how diseases change over time. Meanwhile, his team is still exploring how to apply the work to parts of the body beyond the brain and what types of synthetic images (MRI, CAT, PET) are most useful.

The possibilities are exciting and, like his many roles, “it can be a bit overwhelming,” he said. “There are so many different things we can start thinking about now.”

Comments are closed.