This story was originally published as a news story from Northwestern University McCormick School of Engineering.

A Smarter, More Efficient Way To Read Scans

Physicians and engineers at Northwestern Medicine have created a first-of-its-kind generative AI system that is transforming how radiologists work. Detailed in this study, the tool boosts productivity, identifies serious health problems fast and offers a cost-effective way to address the global radiologist shortage.

“This is, to my knowledge, the first use of AI that demonstrably improves productivity, especially in health care,” says study coauthor Mozziyar Etemadi, MD, PhD, an anesthesiologist at Northwestern Medicine, and assistant professor of Anesthesiology at Northwestern University Feinberg School of Medicine and of Biomedical Engineering at Northwestern University McCormick School of Engineering.  Dr. Etemadi adds that he has not seen gains like this, even in other fields.

Addressing a Growing Radiologist Shortage

Radiology has become one of healthcare’s biggest pressure points. The United States is facing a nationwide radiologist shortage. At the same time, imaging volumes continue to rise by up to 5% each year, while radiology training programs grow at only 2%. 

To help bridge the gap, Northwestern Medicine physicians and engineers developed this AI tool to accelerate report generation without compromising accuracy. In the study, radiologists using the tool cleared backlogs and delivered medical imaging results more quickly than when working without it. And while the AI tool is powerful, experts emphasize that it is designed to assist, not replace, radiologists.

“You still need a radiologist as the gold standard,” says Samir F. Abboud, MD, chief of Emergency Radiology at Northwestern Medicine and clinical assistant professor of Radiology at Feinberg School of Medicine. “Medicine changes constantly — new drugs, new devices, new diagnoses — and we have to make sure the AI keeps up. Our role becomes ensuring every interpretation is right for the patient.”

How the Study Worked

The research team, including Dr. Etemadi, Dr. Abboud and Jonathan Huang, PhD, used and evaluated the AI tool across an 11-hospital Northwestern Medicine network. Nearly 24,000 radiology reports were analyzed over five months. The team compared how long it took to create the reports and their accuracy, with and without AI support.

The results showed that efficiency went up by 15.5% on average, with some radiologists achieving higher gains. Accuracy stayed strong. With the time they saved, radiologists could get diagnoses out faster, especially for urgent conditions.

A First for AI in Clinical Care

According to the research team, this is the first generative AI radiology tool in the world to be integrated into a real clinical workflow. It is also the first time a generative model has shown high accuracy and improved efficiency across all X-ray types, from skull to toe.

Unlike most AI tools that focus on a single condition, the AI model at Northwestern Medicine takes a broader approach. It reviews an entire X-ray or computed tomography (CT) scan and produces a report that is about 95% complete and tailored to the patient and the radiologist’s preferred reporting style. Radiologists then review and finalize the results.

“For me and my colleagues, it’s not an exaggeration to say that it doubled our efficiency,” says Dr. Abboud. “It’s such a tremendous advantage and force multiplier.” 

Spotting Dangerous Conditions Faster

Beyond efficiency, the AI system can identify serious problems, such as pneumothorax or collapsed lung, before a radiologist even opens the image.

As the AI model drafts reports, another automated tool monitors them for critical findings and cross-checks them with patient records. If something serious is detected, the system quickly alerts radiologists.

“On any given day in the Emergency Department, we might have 100 images to review, and we don’t know which one holds a diagnosis that could save a life,” says Dr. Abboud. “This technology helps us triage faster — so we catch the most urgent cases sooner and get patients to treatment quicker.” 

Improving Early Detection

The research team is also adapting the AI model to look for conditions that may be missed or diagnosed late, such as early-stage lung cancer.

“Having a draft report available, even before it is viewed by the radiologist, offers a simple, actionable datapoint that can be quickly and efficiently acted upon,” says Dr. Etemadi. “This is completely different than traditional triage systems, which need to meticulously be trained one by one on each and every diagnosis.”

Built In-House and Designed for Real-World Care

Unlike many commercial medical AI systems, the Northwestern Medicine model was built from scratch rather than using a large internet-trained model. Physicians and engineers developed it based on clinical data from the Northwestern Medicine network. This approach made the generative AI tool faster, more accurate and tailored to radiology needs.

“We’re not just pushing healthcare AI forward,” says Dr. Etemadi. “We’re advancing the fundamentals of AI at a fraction of the cost of the big AI labs.” The study proves it is possible for typical health systems to build custom generative AI models.