Israel’s AI Just Sped Up the MRI Cancer Scan

An MRI is one of medicine’s most powerful cameras, and one of its slowest. A single scan can run half an hour, often too slow for the machine to keep pace with what it is trying to see. Researchers at Israel’s Technion have now found a way to speed it up sharply, and the fix is not a stronger magnet or a more expensive machine. It is mathematics, with a careful dose of AI.

Their method, called ELITE, rebuilds a usable image about once a second, where a conventional dynamic scan manages one every one to two minutes. The work was published in Nature Communications in May, and the Technion presented it to the public on June 22. The target is breast cancer, and the payoff is not only comfort or speed for its own sake. It lets the scan capture something it used to miss.

Dynamic breast MRI works by injecting a contrast dye and watching how tissue takes it up. That pattern is part of the diagnosis: cancerous tissue tends to draw the dye in quickly and release it quickly, while benign tissue usually fills more slowly and steadily. Reading the rhythm well needs fine detail and fine timing at the same moment, and until now the scanner had to trade one for the other. At one image a second, tested on 54 patients, ELITE holds on to both.

Why an MRI takes so long

To see why this is hard, it helps to know what an MRI actually does. It does not take a photograph. It collects raw frequency data, point by point, into a grid called k-space, and only afterward does a computer turn that grid into the image a radiologist reads. Filling the whole grid is what makes a scan slow, stretching a full exam past an hour. For a frightened child, a claustrophobic patient wedged in a narrow tube, or a clinic with a long waiting list, every one of those minutes counts.

The way to go faster is to collect less data and let a computer reconstruct the rest. The danger is that a computer filling in gaps can do more than blur. It can invent. A model trained to supply missing detail might draw in a lesion that was never there, or quietly erase one that is. In a 2020 community challenge, several leading systems did exactly that, and the troubling part was that the fabricated images still scored well on the standard automated quality checks. The conclusion was sober: a human radiologist still has to look.

What ELITE actually does

This is where the Technion team’s design stands out. Instead of handing the whole job to a neural network inventing pixels, ELITE leans first on a mathematical model of how tissue behaves over time, and uses a comparatively small AI network only to clean up noise and the artifacts that undersampling leaves behind. The physics carries the load and the AI tidies up. It is a deliberately conservative approach, which is exactly what you want when the picture decides a diagnosis.

Why it matters

None of this is a lab promise waiting on approval. AI reconstruction is already in hospitals around the world: GE’s deep-learning system cleared the FDA in 2020, with Siemens and Philips shipping their own. What the Technion pushed is the hardest part, the time dimension, where progress had been slow.

The stakes are not abstract. For women at high genetic risk, MRI catches cancers that mammography misses, with sensitivity around 77 percent against roughly 39 percent for the mammogram, though it also raises more false alarms. The limit has rarely been the science. It has been cost and capacity. About 2.3 million women were diagnosed with breast cancer in 2022, and some 670,000 died, most of them where scanners are scarce. A scan that needs less data is a scan that can reach more people.

Where Israel comes in

The advance lands in a country that has quietly become a force in medical-imaging AI. Aidoc, out of Tel Aviv, holds more than thirty FDA clearances and runs in some two thousand hospitals. Ibex does the same for pathology, Nanox.AI for X-ray and CT. By the Israel Innovation Authority’s count, a handful of Israeli companies hold around sixty of the world’s FDA approvals in AI medical imaging, close to six percent of the global total. The Technion itself now ranks first in Europe and among the top two dozen in the world for AI research. The lead author, Eddy Solomon, recently brought his lab to its Faculty of Biomedical Engineering, continuing the work with collaborators at New York University and Weill Cornell. A frontier researcher choosing Haifa is its own quiet headline.

The loudest AI stories this year belong to chatbots and to the models governments want kept behind a fence. The advance that may touch more lives is quieter, and it came out of Haifa: a piece of mathematics that lets a machine keep pace with a tumor instead of trailing it. The frontier does not always look like a frontier. Sometimes it looks like a clearer picture, out of an Israeli lab, arriving in a second instead of two minutes.