A groundbreaking generative AI system developed entirely in-house at Northwestern Medicine is revolutionizing radiology. The system significantly improves productivity, rapidly identifies critical conditions, and offers a promising solution to the global shortage of radiologists. The findings, published in JAMA Network Open, highlight how this tool enhances both speed and diagnostic quality.
Remarkable Gains in Speed Without Compromising Accuracy
Deployed across Northwestern Medicine’s 12 hospitals, the AI analyzed nearly 24,000 radiology reports over five months in 2024. The study showed an average 15.5% improvement in radiograph report completion, with some radiologists reaching up to 40% gains—without any loss in accuracy. Ongoing unpublished work reports up to 80% efficiency gains, extending use to CT scans.
“This is the first AI tool in healthcare to demonstrate real, measurable productivity gains,” said senior author Dr. Mozziyar Etemadi, assistant professor at Northwestern University Feinberg School of Medicine.
Real-Time Integration Into Clinical Workflows
Unlike existing AI tools that focus on specific diseases, this generative model evaluates entire X-rays or CT scans. It then creates a 95% complete, personalized draft report, which radiologists can review and finalize. By doing so, it accelerates diagnoses, especially for time-sensitive cases.
“This system doubled our efficiency. It’s a real force multiplier,” said Dr. Samir Abboud, chief of emergency radiology at Northwestern Medicine.
Immediate Alerts for Life-Threatening Conditions
As reported by medicalxpress, the AI model doesn’t just save time—it can save lives. It flags critical conditions like pneumothorax (collapsed lung) before radiologists even open the file. It also continuously scans for urgent findings and cross-checks them with patient records to issue alerts for immediate action.
Built Independently of Big Tech
Rather than adapting external models like ChatGPT, Northwestern engineers built their AI from scratch using internal clinical data. This approach made the model faster, lighter, and highly specific to radiology needs.
“There’s no need to depend on tech giants. We proved that hospitals can create tailored AI tools themselves,” said first author Dr. Jonathan Huang.
Addressing the Radiologist Shortage
With the U.S. facing a projected shortfall of 42,000 radiologists by 2033, this technology could help reduce backlogs and return results faster. While the AI supports diagnosis, human expertise remains essential.
“Our job is to make sure every interpretation is patient-centered and clinically sound,” added Abboud.
Two patents for this AI system have been granted, with commercialization now underway.