The findings will be published on Thursday (June 5) in JAMA Network Open.
"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, an assistant professor of anesthesiology at Northwestern University Feinberg School of Medicine and of biomedical engineering at Northwestern’s McCormick School of Engineering.
For the study, the AI system was deployed in real-time across the 12-hospital Northwestern Medicine network, where nearly 24,000 radiology reports were analyzed over a five-month period in 2024. Etemadi’s team then compared radiograph report creation times and clinical accuracy with and without the AI tool.
The results: an average 15.5% boost in radiograph report completion efficiency - with some radiologists achieving gains as high as 40% - without compromising accuracy. Follow-on work, still unpublished, shows up to 80% efficiency gains and enables the tool for CT scans. The time saved allowed radiologists to return diagnoses much faster, particularly in critical cases in which every second counts.
According to the study authors, this is the first generative AI radiology tool in the world to be integrated into a clinical workflow. It also is the first time a generative AI model has demonstrated both high accuracy and increased efficiency across all types of X-rays, from skulls to toes.
Unlike other narrow AI tools currently on the market that focus on detecting a single condition, Northwestern’s holistic model analyzes the entire X-ray or CT scan. It then automatically generates a report that is 95% complete and personalized to each patient, which the radiologist can choose to use, review and finalize. These reports summarize key findings and offer a template to augment the radiologists’ diagnosis and treatment.
"For me and my colleagues, it’s not an exaggeration to say that it doubled our efficiency. It’s such a tremendous advantage and force multiplier," said co-author Dr. Samir Abboud, chief of emergency radiology at Northwestern Medicine and clinical assistant professor of radiology at Feinberg.
In addition to improving efficiency, the AI system flags life-threatening conditions like pneumothorax (collapsed lung) in real time - before a radiologist even looks at the X-rays. As the AI model drafts reports for every image, an automated tool monitors those reports for critical findings and cross-checks them with patient records. If the system identifies a new condition that needs urgent intervention, it could immediately alert radiologists.
"On any given day in the ER, we might have 100 images to review, and we don’t know which one holds a diagnosis that could save a life," Abboud said. "This technology helps us triage faster - so we catch the most urgent cases sooner and get patients to treatment quicker."
The Northwestern team also is adapting the AI model to detect potentially missed or delayed diagnoses, such as early-stage lung cancer.
Rather than adapting large, internet-trained models like ChatGPT, the Northwestern engineers built their own system from scratch using clinical data from within the Northwestern Medicine network. That allowed the team to create a lightweight, nimble AI model designed specifically for radiology at Northwestern - faster, more accurate and requiring far less computing power.
"There is no need for health systems to rely on tech giants," said first author Dr. Jonathan Huang, a third-year medical student at Feinberg who holds a Ph.D. in biomedical engineering from McCormick.
"Our study shows that building custom AI models is well within reach of a typical health system, without reliance on expensive and opaque third-party tools like ChatGPT. We believe that this democratization of access to AI is the key to drive adoption worldwide," Etemadi added.
Radiology is becoming one of health care’s biggest bottlenecks. By 2033, the U.S. is expected to experience a shortage of up to 42,000 radiologists, as imaging volumes rise by up to 5% annually while radiology residency positions increase by just 2%.
Northwestern's AI system offers a solution, helping radiologists clear backlogs and deliver results in hours instead of days. And while the technology is powerful, it won't replace humans.
"You still need a radiologist as the gold standard," Abboud said. "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."
Two patents have been approved for the Northwestern Medicine technology and others are in various stages of the approval process. The tool is in the early stages of commercialization.
Huang J, Wittbrodt MT, Teague CN, Karl E, Galal G, Thompson M, Chapa A, Chiu ML, Herynk B, Linchangco R, Serhal A, Heller JA, Abboud SF, Etemadi M.
Efficiency and Quality of Generative AI-Assisted Radiograph Reporting.
JAMA Netw Open. 2025 Jun 2;8(6):e2513921. doi: 10.1001/jamanetworkopen.2025.13921