The dataset is part of fastMRI, a collaborative research effort that NYU School of Medicine launched with Facebook in August. Under the collaboration, the medical school’s radiology department will work with the social network’s Facebook AI Research division to apply artificial intelligence to MRI scans.
The fastMRI project centers on changing the way an MRI machine creates an image. Today, an MRI puts together a rendition of a patient’s internal organs by gathering a series of sequential views of a patient, which it translates into a cross-sectional image. This is a time-consuming process, which makes MRIs challenging for children, elderly patients and those with claustrophobia.
However, early research conducted at NYU School of Medicine suggests AI can generate high-quality images from limited data, meaning an MRI machine may be able to capture less data — and therefore scan more quickly — and still create an accurate image. In fact, NYU School of Medicine and Facebook think that, with AI, they can complete MRI scans up to 10 times faster.
The first phase of fastMRI focuses on using AI to complete MRI scans of the knee, with the initial data release comprising more than 1.5 million knee images pulled from 10,000 MRI scans. The data, which was collected by researchers NYU School of Medicine, has been fully anonymized and stripped of patient identifiers.
“We hope that the release of this landmark data set … will provide researchers with the tools necessary to overcome the challenges inherent in accelerating MR imaging,” said Michael Recht, MD, chair of the radiology department at NYU Langone Health in New York City. “This work has the potential to not only help increase access to MR imaging, but also improve patient care worldwide.”
To encourage other researchers to contribute to the project, Facebook has released a set of open-source AI models and tools alongside the dataset.
In an August blog post, Facebook highlighted that its participation in the project is limited to supplying the fastMRI team with its expertise in AI and computer vision. The fastMRI project does not use any Facebook data.
“FastMRI not only could have an important impact in the medical field, it’s also an interesting research challenge that will help to advance the field of AI,” said Larry Zitnick, PhD, research lead at Facebook AI Research. “The dataset NYU Langone is releasing and the baseline models we’ve open-sourced will enable other researchers to join us in working on this challenging problem, and we believe this open approach will bring about positive results more quickly.”
To request access to the fastMRI dataset, click here.
More articles on artificial intelligence:
Amazon adds 3 ‘HIPAA-eligible’ AI tools to cloud service
CMS to launch AI challenge
Northwestern introduces AI tool to help medical assistants conduct sonograms
