NVIDIA’s Cambridge-1 Supercomputer And MONAI Were Utilized By Researchers At King’s College London To Develop Open-Source Synthetic Brain Pictures, Which Will Help To Speed Up AI In Healthcare

This Article is written as a summay by Marktechpost Staff based on the Research Article 'The Man With 100,000 Brains: AI’s Big Donation to Science'. All Credit For This Research Goes To The Researchers of This Project. 

Please Don't Forget To Join Our ML Subreddit

King’s College London, in collaboration with partner hospitals and universities, revealed new details about one of the first projects on Cambridge-1, the UK’s most powerful supercomputer.

Cambridge-1 is speeding up health research in medical imaging, genetics, and medication development. Cambridge-1 is dedicated to advancing UK health research through digital biology, unlocking a deeper understanding of disease and breakthroughs in medicine, with its founding partners — AstraZeneca, GSK, Guy’s and St Thomas’ NHS Foundation Trust, King’s College London, and Oxford Nanopore — and other UK organizations.

The Synthetic Brain Project aims to develop deep learning models that can create 3D MRI pictures of human brains from scratch. These models can aid scientists in gaining a better understanding of the human brain across a range of ages, genders, and disorders. The AI models were created by King College London’s and NVIDIA data scientists and engineers as part of the UK Research and Innovation-funded London Medical Imaging & AI Centre for Value-Based Healthcare research and a Wellcome Flagship Programme (in collaboration with University College London).

The AI models were created to diagnose neurological ailments based on brain MRI scans. Still, they might also forecast diseases that a brain might acquire over time and allow for preventative therapy. The use of synthetic data has the added benefit of ensuring patient anonymity. The photos were created and allowed King College London to share the findings with the rest of the UK healthcare community. Without Cambridge-1, the AI models would have taken months to train instead of weeks, and the image quality would have been less clear. Cambridge-1 was utilized by King’s and NVIDIA researchers to scale the models to the required size using several GPUs. Then, a procedure known as hyperparameter tweaking was employed to significantly increase the models’ accuracy.

Science using Synthetic Data

The photos represent a new branch of synthetic data in healthcare, which is already widely employed in consumer and corporate computer vision programs. Those areas, ironically, have access to accessible databases, including millions of real-world photos.

On the other hand, medical photos are uncommon, often available only to researchers affiliated with big institutions due to the necessity to safeguard patient privacy. Even so, medical pictures tend to represent the demographics of the hospital’s patients rather than the general public.

Female and male brains, aged and young brains, diseased and unaffected brains. The new AI technique has the advantage of creating pictures on demand. It makes them once you plug in what you need. Although synthetic, the images are precious because they keep crucial biological properties, making them seem and operate precisely like actual brains.

On Cambridge-1, scaling with MONAI.

The task necessitated the use of a supercomputer with super software. The engine was NVIDIA Cambridge-1, a supercomputer dedicated to groundbreaking AI research in healthcare. The program was powered by MONAI, an AI platform for medical imaging.

They collaborated to construct an AI factory for synthetic data that allowed researchers to perform hundreds of trials, choose the top AI models, and generate images using inference.

“Without Cambridge-1 and MONAI, we wouldn’t have been able to do this job,” the researcher added.

Images with up to 10x speedups

Cambridge-1 is an NVIDIA DGX SuperPOD with 640 NVIDIA A100 Tensor Core GPUs, each with enough RAM to process one or two of the team’s gigantic 16 million 3D pixel pictures.

Domain-specific data loaders, metrics, GPU-accelerated transformations, and an efficient workflow engine are among MONAI’s building pieces. According to the researcher, the software’s clever caching and multi-node scalability may speed up processes up to 10 times.

Beyond the Mind’s Eye

The 100,000 brain pictures will be hosted by Cardoso in collaboration with Health Data Research UK, a national repository. Researchers will also have access to AI models, allowing them to build any pictures they want.

There’s more to come. The researchers are looking at how the models may be used to create 3D pictures of any portion of the human body in any medical imaging method, including MRIs, CAT scans, and PET scans.

Conclusion:

The Synthetic Brain Project intends to produce deep learning models capable of producing 3D MRI images of human brains from scratch. These models can help scientists learn more about the human brain related to different ages, genders, and illnesses. The researcher lauded the study for pointing in several directions as if it were releasing the contents of other minds. Using synthetic images, researchers will be able to evaluate how diseases grow over time. Meanwhile, the team determines how to extend their results to sections of the body other than the brain and which types of synthetic imaging (MRI, CAT, and PET) are most successful.