AI cuts average CCTA reading time by 73%, helping radiologists detect coronary artery disease

Artificial intelligence can significantly reduce post-processing and interpretation times for coronary CTA scans and help radiologists detect a particularly deadly type of cardiovascular disease, according to a new report.

Imaging and heart experts developed their deep learning-powered tool using data from 165 patients. The software automatically reformats CCTA images to identify abnormalities within major arteries contributing to obstructive coronary artery disease.

AI reduced the average required reading time by 73% compared to human-only interpretations, the authors reported Saturday in the European Journal of Radiology. The tool was more diagnostically accurate than inexperienced rads and comparable to intermediate and advanced imaging experts.

“Thus, AI software as a second reader may substantially shorten reading time, potentially improve the diagnostic accuracy of inexperienced readers, and increase the confidence of more experienced readers,” Chun Yu Liu, with Jinling Hospital’s Department of Diagnostic Radiology in Nanjing, China, and colleagues wrote.

Coronary artery anatomy is complex and assessing stenosis is particularly challenging. Training and experience are both key factors in accurately interpretating CCTA exams, the authors added.

With this in mind, Liu et al. sought to automate the process. They assessed the performance of AI, human reading alone, and pairing AI with cardiovascular radiologists. Invasive coronary angiography was used as reference standard and CAD was defined as luminal stenosis of 50% or greater.

On a per-patient level, AI diagnosed obstructive CAD with 90.5% sensitivity, 82.3% specificity and notched an area under the curve score of 0.90. Rads’ post-processing time was decreased from 8.5 minutes to 3.7 minutes with the help of AI.

And given image quality and arterial calcification did not alter the performance of the software, Liu and co-authors see real-world clinical potential for their tool

“Importantly, image quality and calcification burden did not negatively affect the diagnostic performance of AI,” they added. “Thus, AI may be an efficient tool for the diagnosis of coronary artery stenosis.”

Read the full study here.

Around the web

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.

Parts of the device can move or become loose during procedures, the FDA warned. 

Trimed Popup
Trimed Popup