The first in-house developed in-house at Northwest Medicine will revolutionize radiology, increase productivity, identify life-threatening conditions in milliseconds, and provide a breakthrough solution to the global shortage of radiologists, a large-scale new study has discovered.
“This is, to my knowledge, the first use of AI to clearly improve productivity, especially in healthcare. Even in other areas, I have never seen a boost of nearly 40%,” says Mozziyar Etemadi, assistant professor of anesthesiology at the Feinberg School of Medicine of Medicine of Biomedical Engineering of Biomedical Engineering, Northwest University.
For this study, the AI system was deployed in real time across the 11 Hospitals Northwest Medical Network, where approximately 24,000 radiation reports were analyzed in 2024 over five months. Etemadi's team then compared radiograph production times and clinical accuracy, with or without AI tools.
Results: An average of 15.5% of X-rays has been reported. Some radiologists have achieved 40% higher profits. The yet-unpublished follow-on work introduces efficiency gains of up to 80% and enables the CT scan tool. The time saved allowed radiologists to return the diagnosis faster, especially in critical cases counted per second.
According to the study authors, it is the world's first generation AI radiology tool integrated into clinical workflows. It is also the first time that a generative AI model has demonstrated both high accuracy and increased efficiency across all types of x-rays, from the skull to the toes.
“It doubled our efficiency.”
Unlike other narrow AI tools currently on the market, Northwestern's holistic models analyze an entire x-ray or CT scan, as they focus on detecting a single condition. It then automatically generates 95% complete and personalized reports for each patient, in the radiologist's own reporting style that the radiologist can choose to use, review and finalize. These reports summarise important findings and provide templates to enhance radiologist diagnosis and treatment.
“To me and my colleagues, it's not an exaggeration to say it doubled our efficiency. It's a huge advantage, it's a forced multiplier,” said Samir Abdo, Head of Emergency Radiology at Northwest Medicine and Assistant Professor of Clinical Radiology at Feinberg.
https://www.youtube.com/watch?v=wor80psraom
Flag life-threatening conditions
In addition to improving efficiency, AI systems also flag life-threatening conditions like pneumothorax (folding lungs) in real time before radiologists see x-rays.
Just as AI models draft reports for all images, automated tools monitor these reports for important discoveries and cross-check them with patient records. If the system identifies a new condition requiring urgent intervention, it may immediately alert the radiologist.
“There may be 100 images to review at any time in the ER, but I don't know which one can save lives,” Abboud says. “This technology helps triage faster, so it catches the most urgent cases faster and treats patients more quickly.”
The Northwestern team is also adapting AI models to detect potentially missed or delayed diagnoses, such as early stage lung cancer.
“Even before being seen by a radiologist, making draft reports available provides simple and practical data points that allow you to act quickly and efficiently, completely different from traditional triage systems where you need to train closely one by one with each diagnosis.
“There's no need to rely on high-tech giants.”
Rather than adapting large-scale internet-trained models like CHATGPT, Northwestern Engineers built their own systems from scratch using clinical data within the Northwestern Medicine Network. This allowed the team to create lightweight, agile AI models specifically designed for Northwestern radiology.
We are not just moving forward with healthcare AI. It advances the fundamentals of AI at the cost of some of the big AI labs. This is the beginning of the Deepseek moment in Healthcare AI.
“We don't need a healthcare system to rely on the tech giant,” said Jonathan Fan, a third-year medical student at Feinberg, who holds a PhD in biomedical engineering at McCormick.
“Our research shows that building custom AI models is within the scope of typical healthcare systems without relying on expensive, opaque third-party tools like ChatGpt. We believe this democratization of access to AI is key to driving adoption around the world,” Etemadi added.
Etemadi led a Belllab-style engineering team built into the hospital system, attracting the talent of top big technology and finance.
“My most proud achievement is building a strong, interdisciplinary team that can carry out the top priorities of the healthcare system,” Etemadi said. “We're not just moving forward with healthcare AI. We're moving forward with the foundations of AI at the cost of some of the big AI labs. This is the beginning of the Deepseek moment in Healthcare AI.”
Address global bottlenecks
Radiology is becoming one of the biggest bottlenecks in healthcare. By 2033, the US is expected to experience a shortage of up to 42,000 radiologists as imaging volumes increase by 5% per year and radiology residences increase by just 2%.
Northwestern's AI systems provide solutions and help radiologists clear their backlog and deliver results in hours rather than days. And while this technology is powerful, it does not replace humans.
“We still need radiologists as the gold standard,” Abud said. “We must constantly change drugs, new drugs, new devices, new diagnoses, and make sure AI is maintained. Our role is to ensure that all interpretations are correct for our patients.”
Two patents for Northwest Medical Technology have been approved, with the other patents at various stages of the approval process. This tool is in the early stages of commercialization.
