AIRS Medical Wins ‘fastMRI Challenge’ Hosted by Facebook AI Research & NYU Langone Health


AIRS Medical, a Korean startup specializing in medical artificial intelligence (AI) software, has won “fastMRI Challenge 2020,” a global competition jointly organized by Facebook AI Research and NYU Langone Health. The challenge was designed to accelerate Magnetic Resonance Imaging (MRI) scans of the brain using AI. AIRS Medical came in first in all competitive categories — Multi-Coil 4X, Multi-Coil 8X, and Transfer. The winning South Korean AI company presented its accomplishments at ‘Medical Imaging Meets NeurIPS (Neural Information Processing Systems) Workshop 2020’ held on Dec. 11. NeurIPS is an internationally renowned machine learning conference.

‘fastMRI’ is a collaborative research project between Facebook AI Research and NYU Langone Health. It was launched to investigate the use of AI to make MRI scans faster. The fastMRI Challenge is an annual competition open to the global research community in the field of AI and MRI. Participants from research institutions, academia and industry fields around the world competed in this year’s challenge, submitting reconstructed brain MRI images accelerated by factors of 4 and 8.

Submissions were evaluated in both quantitative and qualitative ways, first by SSIM (Structural Similarity Index Measure) and then by radiologists who focused on pathological accuracy and clinical usefulness.

AIRS Medical posted the highest SSIM scores in the four most common MRI sequences. It was also rated the highest by radiology professionals from Mayo Clinic, Stanford University and many other renowned medical institutions in the USA and Europe, who were asked to rank the submissions according to “overall quality of depiction of pathology compared against the gold standard.”

MRI image reconstruction has been a field long dominated by major MRI vendors like Siemens and GE. The emergence of AI-based image reconstruction in medical imaging from the mid-2010s, however, has allowed more players to join the competition including AI researchers and startup companies for faster yet accurate MRI scans.

The key technology in AI-based image reconstruction lies in producing accurate images from under-sampled MRI data by utilizing well-trained algorithms. The participation of major MRI vendors in this challenge as well as consecutive AI-based software introductions suggests that this new technology could soon come to replace conventional methods of acceleration.

“The demand for diagnostic imaging including MRI scans never stops growing as healthcare systems in both developed and developing countries tend to put more weight on prevention and early diagnosis of diseases. It is also true that MRI is a difficult market for new players because of its high technology barriers to entry. The fact that a Korean startup won in a competition dealing with medical imaging technology, traditionally a field where just a few developed countries used to dominate, shows AIRS Medical’s competency in the global market” says Lee Jin-gu, CEO of AIRS Medical. “Also, AIRS was ranked top in the qualitative evaluation by radiologists, our potential customers, which is really meaningful for us. We are very confident and excited to launch our very first AI-augmented solution next year,” he added.

AIRS Medical is a Seoul-based AI medical startup founded by a group of Seoul National University graduates. The company was chosen as one of the 40 innovative startups in the “Baby Unicorn 200 Fostering Project” organized by the Ministry of SMEs and Startups (MSS) this June. Its first commercialized product is expected to get clearance from the U.S. FDA and the Korean Ministry of Food and Drug Safety (KMFDS) next year.