AI software helps radiologists cut unnecessary chest CT scans by 30%

Radiologists using a newly developed deep learning algorithm can more accurately detect lung cancers on X-rays, helping them properly recommend follow-up CT scans to avoid unnecessary imaging.

The study was undertaken by Massachusetts General Hospital and Seoul, South Korea-based Lunit, which developed the artificial intelligence software. During testing, eight readers improved their ability to spot patients without cancer by 30%, the team reported in European Radiology.

These gains translated into more appropriate CT orders, particularly for younger rads, and 30% fewer unnecessary scans.

"The use of AI could help to detect pulmonary nodules accurately with chest X-rays, as well as reduce the need for unnecessary chest CT exams in some patients," co-author Mannudeep K. Kalra, MD, a radiologist at MGH, said Thursday. "This finding can benefit patients by enabling them to avoid unneeded radiation exposure, and it can benefit the healthcare system by preventing certain medical costs."

The tool was trained using more than 3 million pieces of medical data and is currently used in more than 300 hospitals to detect major chest diseases, including lung nodules and tuberculosis. It’s expected to gain FDA approval sometime this year.

For their study, Kalra et al. selected 519 patient scans from the National Lung Screening Trial. Three radiology residents and five board-certified rads interpreted the images with AI and without it.

With assistance, residents recommended 28% more chest CTs (54.7% vs. 70.2%) for patients with visible lung cancer. And radiologists recommended nearly 30% fewer unnecessary scans for cancer-negative individuals (16.4% vs. 11.7%).

Overall, the software can help doctors diagnose patients with enhanced efficiency, spotting potential cancers earlier while saving time and costs for those who don’t require follow-up, the authors explained.

Two researchers listed in the study are employed by Seoul, South Korea-based Lunit.

Read the full study here (behind paywall).

Around the web

The difference when using speech-recognition software may be accentuated in certain groups of radiologists, UNC researchers detailed in the Journal of Digital Imaging.

Researchers have used machine learning to track diabetes at the population level.

AI developers have worked with experts in human-computer interaction to design an EHR that shows clinicians all information pertinent to the patient case they’re working on—and only that info.

Trimed Popup
Trimed Popup