AI Halves MRI Exam Time

Magnetic resonance imaging works by collecting raw signals from hydrogen atoms and turning those signals into pictures, so every extra slice usually costs extra seconds inside the scanner. That is why a routine magnetic resonance imaging exam can feel less like a photo and more like a long audio recording where the machine has to capture enough data before it can play the image back. (ajnr.org) The bottleneck has always been time. The American Journal of Neuroradiology said long magnetic resonance imaging acquisition times raise costs, make patients uncomfortable, and create motion artifacts, which are the blurs caused by breathing, swallowing, or shifting even a little during the scan. (ajnr.org) Older speed-up tricks already existed, including parallel imaging and compressed sensing, but they often traded away signal quality or introduced new artifacts when radiologists pushed them too far. Newer deep learning reconstruction systems try to keep the speed gain while cleaning up the missing or noisy parts of the image afterward. (ajnr.org) The basic idea is simple: scan less, then let software rebuild what a full scan would probably have shown. NYU Langone described this as using artificial intelligence to reconstruct missing information from rapid scans, after training on large sets of conventional magnetic resonance imaging data. (nyulangone.org) That approach has now moved from research papers into day-to-day hospital workflow. Antoni van Leeuwenhoek Hospital in Amsterdam told Radiology Business that abdomen magnetic resonance imaging exams that used to take about 23 minutes now finish in about 9 minutes with on-scanner software that fills in gaps between slices using synthetic images. (radiologybusiness.com) The gain is not just a shorter appointment on the calendar. The Amsterdam team said the department now scans 18 additional patients per week, which means more people get imaged sooner without opening extra evening or weekend slots. (radiologybusiness.com) Shorter scans also fix one of magnetic resonance imaging’s oldest enemies: motion. In the Amsterdam report, radiologist Doenja Lambregts said the synthetic-image workflow can produce equal or better image quality partly because patients spend fewer minutes breathing and moving while the scanner is collecting data. (radiologybusiness.com) Other hospitals are seeing the same pattern across different body parts. At the Radiological Society of North America meeting in December 2025, a multisite network in Argentina reported mean scan-time reductions of 47.9% after adding artificial intelligence reconstruction to five scanners, with shoulder exams down 64% and ankle exams down 63.9%. (rsna.org) Vendors are now selling this as a scanner feature, not a lab demo. Siemens Healthineers says its Deep Resolve Boost software can deliver up to a 73% scan-speed increase when combined with parallel imaging and simultaneous multi-slice methods, which is why hospitals are starting to treat reconstruction software like capacity equipment. (siemens-healthineers.com) The catch is that faster pictures still have to be trustworthy pictures. The American Journal of Neuroradiology review published in January 2026 said closed-source vendor algorithms, scanner-to-scanner generalizability, and the risk of changing how disease appears on the image are still open questions for widespread adoption. (ajnr.org) So the story is not that artificial intelligence is replacing the radiologist staring at the scan. The story is that artificial intelligence is increasingly replacing part of the waiting time, which is why a machine that used to behave like a one-lane road is starting to act more like a multi-lane highway. (radiologybusiness.com)