Magnetic resonance imaging, the gold standard of modern diagnostic imaging, provides doctors with greater detail than other imaging techniques, such as X-ray or CT scans. But MRI scans take a long time—anywhere from 15 minutes to an hour or more. All the while, the patient must lie still inside a claustrophobia-inducing metal tube.

That scenario could change. Facebook’s Artificial Intelligence Research (FAIR) group and the New York University School of Medicine recently announced a collaborative research project that will use machine-learning techniques to make MRI scans up to 10 times faster.

“Using AI, we believe it may be possible to capture less data and therefore image faster, while still preserving or even enhancing the rich information content of MR images,” Dr. Daniel Sodickson, vice chair for research in radiology at NYU School of Medicine, told Forbes.

Facebook researchers will train their “fastMRI” model using an NYU-provided data set of 3 million images of the knee, brain, and liver. The challenge, according to a Facebook blog, is to get the artificial intelligence network to “recognize the underlying structure of the images in order to fill in views omitted from the accelerated scan” and to bridge those gaps without sacrificing accuracy. Missing a few key pixels could mean the difference between a clear scan and one showing an anomaly such as a torn ligament or a tumor.

If the project is successful, the researchers believe the benefits of faster MRIs would extend beyond a more comfortable patient experience.

“You also get increased accessibility in areas with MRI shortages and you can get improved image quality when you’re trying to image things that move fast, like the heart,” Sodickson told Forbes. “If we can get it fast enough to replace X-rays or CT [scans] then we can also reduce radiation exposure for the population while still getting the critical medical information.”

koya79/iStock

Artificial Shakespeare

Artificial intelligence (AI) algorithms called artificial neural networks have generated original music and art, some of it to critical acclaim. But until now, few researchers have tried to construct an AI system that writes poetry.

In a paper presented at the 2018 meeting of the Association of Computational Linguistics, researchers described their deep-learning algorithm called “Deep-speare,” designed to create poems that match the style and beauty of Shakespeare’s sonnets.

Using a data set of 2,685 English-language sonnets, the researchers trained the AI system to generate four-line poems. These poems were randomly given, along with human-authored poems, to human judges who did not know the source of each poem.

The AI poems scored high on form, such as meter and rhyme, but fell short on readability and emotion, according to the researchers. Here’s an example of a poem generated by Deep-speare: “Shall I behold him in his cloudy state / for just but tempteth me to stop and pray / a cry: if it will drag me, find no way / from pardon to him, who will stand and wait.” —M.C.