AI Radiology Tool Identifies Life-Threatening Conditions in Milliseconds

By MedImaging International staff writers
Posted on 11 Jun 2025

Image: Samir F. Abboud, MD, chief of emergency radiology at Northwestern Medicine, and co-author of the study detailing the new generative AI tool for radiology (Photo courtesy of José M. Osorio/Northwestern Medicine)

Radiology is emerging as one of healthcare’s most pressing bottlenecks. By 2033, the U.S. could face a shortage of up to 42,000 radiologists, even as imaging volumes grow by 5% annually. Researchers have now developed the first generative artificial intelligence (AI) system tailored for radiology, aiming to dramatically enhance productivity, accuracy, and speed.

Unlike existing AI tools that focus on detecting a single condition, this comprehensive model developed at Northwestern University (Evanston, IL, USA) analyzes full X-rays or CT scans and automatically drafts 95% complete, patient-specific reports. These reports are designed to support radiologists in making faster, more informed decisions and can be edited and finalized as needed.

The system also flags life-threatening conditions like pneumothorax in milliseconds—before the radiologist has even reviewed the scan. A built-in monitoring tool cross-checks AI-generated reports with patient records and immediately alerts radiologists to critical, new conditions requiring urgent intervention. What sets this AI apart is that it was trained entirely on real-world clinical data from the Northwestern Medicine network—unlike many models adapted from internet-trained tools like ChatGPT. This allowed the team to create a streamlined, high-performance model that is faster, more accurate, and requires significantly less computing power.

In a study published in JAMA Network Open, the AI system was tested across 12 Northwestern hospitals, where it analyzed nearly 24,000 radiology reports over five months in 2024. The results showed an average 15.5% increase in radiograph report efficiency, with some radiologists improving by as much as 40%—all without sacrificing diagnostic accuracy. Follow-up testing (yet to be published) has shown up to 80% gains and extended the tool’s use to CT scans.

By accelerating report turnaround times and flagging critical findings earlier, the AI helps radiologists clear backlogs and return diagnoses in hours instead of days. According to the team, this is the first generative AI radiology tool integrated into a live clinical workflow—and the first to show both high accuracy and efficiency gains across all X-ray types. Work is already underway to expand the system to detect subtle or delayed diagnoses, such as early-stage lung cancer.

“This is, to my knowledge, the first use of AI that demonstrably improves productivity, especially in health care. Even in other fields, I haven’t seen anything close to a 40% boost,” said senior author Dr. Mozziyar Etemadi.

Related Links:
Northwestern University