The FastMRI project is a 2 year-old collaboration between NYU Langone Health and Facebook.

This new paper shows that the time a patient must spend in an MRI machine can be divided by 4 with no degradation in image quality. 1/N

Blog post: https://ai.facebook.com/blog/fastmri-breakthrough-shows-ai-accelerated-mris-interchangeable-with-slow-traditional-mris

#AI4good

4x-subsampled K-space data is collected and fed to an architecture dubbed End-To-End Variational Network, which iteratively applies a U-Net-style Convolutional Net to progressively refine the reconstructed image.

Underlying method described here: https://arxiv.org/abs/2004.06688 

2/N

MRI exams that used to take 1 hour can now be done in 15 minutes: faster diagnosis, less aggravation, less cost.

Data and models are open source.
Paper: https://www.ajronline.org/doi/abs/10.2214/AJR.20.23313

Absolutely fabulous work!
Congrats to the entire team!
@DanielSodickson

3/N, N=3.