Artificial intelligence software to help radiologists find problems earlier

February 23, 2021
An image of the heart as it shows up on the artificial intelligence radiology software.
AI-Rad Companion provides 3D visualizations as well as detects and quantifies calcifications in the coronary arteries that supply the heart with blood. Image provided

An artificial intelligence program for radiologists proved so successful during research trials conducted at the Medical University of South Carolina that the technology will now be used across the MUSC Health system whenever a patient receives a chest CT scan.

It’s a tangible outcome of the strategic value partnership between MUSC and Siemens Healthineers, which blends the medical technology prowess of Siemens Healthineers with the clinical and research expertise at MUSC to refine technologies to the benefit of patients. In this case, the artificial intelligence, called AI-Rad Companion, saved time for radiologists analyzing complex scans and also showed that it could detect abnormalities that might otherwise have gone unnoticed because they were unrelated to the reason the patient needed a CT scan.

AI-Rad Companion can identify widened aortas, coronary artery calcification and vertebral height measurements, which can signal a vertebral fracture. The software integrates into the imaging infrastructure already in place. This means that the hospitals don’t need new or different CT machines, so images taken at hospitals in MUSC Health’s Regional Health Network will also get the benefit of the artificial intelligence, and radiologists don’t need to toggle between different computer programs to review the results.

Sanford Zeigler, M.D., director of thoracic aortic surgery at the MUSC Health Aortic Center, said this early detection benefits patients.

“The vast majority of thoracic aortic aneurysms are asymptomatic right up to the point that they become life threatening, but if an aneurysm is picked up early, medical management can drastically lower the risk that they progress to a more life-threatening problem,” he said.

“Technology like this will help us keep patients safe by avoiding missed diagnoses, and it will help researchers better understand the natural history of this difficult-to-diagnose disease. Additionally, the standardized protocol and landmark-based measurements will allow doctors to more closely track changes over the entire aorta with greater consistency,” Zeigler continued.

U. Joseph Schoepf, M.D., director of cardiovascular imaging for MUSC Health and assistant dean for clinical research in the Medical University of South Carolina College of Medicine, is a longtime collaborator with Siemens Healthineers. In previous work, he studied how well the company’s AI-Rad Companion program could analyze lung problems in emphysema patients.

An image of diseased lungs as they show up on the artificial intelligence radiology software 
The pulmonary module of AI-Rad Companion can detect and measure the size of abnormal lesions in the lung as well as the amount of damaged air sacs, which causes emphysema. Image provided

Schoepf also led the team whose work solidified MUSC Health’s decision to use AI-Rad Companion with all chest CTs. The team, including chest radiologists Dhiraj Baruah, M.D., Jeremy Burt, M.D., and Ismail Kabakus, M.D., ran chest CT scans through the AI-Rad Companion software. At about 50 to 70 scans performed at MUSC Health Charleston each day, they have at this point run thousands of scans through the software.

The researchers looked at the scans of all adult patients, regardless of the reason the scan was ordered. Some patients were suspected to have pneumonia, others were being scanned for lung cancer and in still other cases, doctors were looking for blood clots in the lungs. Schoepf said that whereas humans tend to home in on the cause of the immediate problem, the artificial intelligence analyzes the scans holistically, pointing out all possible problems.

“I basically start looking to answer that clinical question,” Schoepf explained of the radiologist’s role. “But other things in the scan often times fall by the wayside because we’re focusing on a discrete clinical question and can potentially miss things that are technically visible but are not on our radar.”

The researchers determined that not only did the software flag problems that might have been overlooked, but it enabled radiologists to get more done. Schoepf’s team was able to analyze thoracic images 20% faster, meaning radiologists can analyze more scans and help more patients.

Schoepf’s team has also pivoted to the COVID-19 pandemic to see how this technology might help. He said the team is currently studying how the artificial intelligence could help to determine which hospitalized COVID-19 patients are likely to have more severe disease, based on the state of their lungs.

“We're trying to identify patients who would benefit from more intense treatment. Especially as our armamentarium of COVID treatment increases, with more antibodies and pharmaceutical therapy, I believe it would help us to target those therapies better and ensure those patients who really need them, get them,” he said.