The use of AI in thoracic imaging has begun to demonstrate “cumulative evidence of effectiveness,” but more testing and research are needed to determine its feasibility for this application, according to a commentary published February 25 in Radiology.
“Practical AI implementation will require objective onsite performance evaluation, institutional information technology infrastructure integration, and postdeployment monitoring,” wrote a team led by Eui Jin Hwang, MD, PhD, of Seoul National University Hospital in South Korea.
AI for thoracic imaging includes using it for reading chest radiographs and low-dose chest CT scans for lung cancer screening and for triaging pulmonary embolism on chest CT scans, the group noted. Other potential uses researchers are investigating are filtering out normal chest radiographs, monitoring reading errors, and automated opportunistic screening of incidentally identified disease.
But implementing the technology in a daily radiology workflow will require establishing particular measures, such as “educating radiologists and radiology trainees, alleviating liability risk, and addressing potential disparities due to the uneven distribution of data and AI technology,” Hwang and colleagues explained.
Next-generation AI technology shows promise, however.
“[Large] language models (LLMs), including multimodal models, which can interpret both text and images, are expected to innovate the current landscape of AI in thoracic radiology practice,” the authors wrote. “These LLMs offer opportunities ranging from generating text reports from images to explaining examination results to patients.”
AI technology continues to be developed, and may “ultimately prove to be highly useful in diagnostic thoracic imaging and clinical decision support for clinicians,” wrote Edwin J.R. van Beek, MD, of the University of Edinburgh in the U.K. in an accompanying editorial.
“Proper validation and proof of effectiveness, ideally with large randomized studies comparing standard evaluation with the standard evaluation in addition to AI, will make the implementation more likely,” he noted.
The complete commentary can be found here.