How AI is speeding up MRIs: An interview with Nafissa Yakubova, fastMRI project lead

By Mike Schroepfer
September 9, 2020

Facebook AI and our partners at NYU Langone Health are using AI to dramatically speed up the MRI scanning process. The project, called fastMRI, has passed an important milestone: we've demonstrated, for the first time, that AI-generated MRI images created with four times less data are diagnostically interchangeable with traditional MRIs. This means expert radiologists came to the same diagnostic conclusion from the speed up MRIs as the originals.

Recommended Reading

To do this, in partnership with NYU Langone, we created the largest, open-sourced de-identified MRI data set and we used it to train a neural network which enables the AI to generate images from less data. We're continuing to test this on more use cases and other types of MRI machines, but this is an exciting milestone that could help make MRIs faster for patients everywhere. Faster scans could broaden access to MRIs, especially in areas of the world where there are few machines, and they’ll be much better for pediatric patients, who often need to be anesthetized for scans today.

FastMRI is a great example of how our open, collaborative approach to research is driving innovation. By bringing together experts in AI and medical imaging, open-sourcing our models, publishing our work, and sharing the fastMRI datasets, we believe this project will one day soon make a positive impact in millions of people’s lives.

I spoke with Nafissa Yakubova, Facebook’s project lead for fastMRI, to hear more about the challenges the teams overcame and what comes next.