AI is most beneficial as a prescreening tool, new research suggests
Artificial intelligence-enabled applications have been tested in multiple radiology settings, but a new analysis suggests they may be most useful as prescreening tools.
CT lung cancer screenings have been identified as one of numerous areas where AI assistance could enhance radiologists’ accuracy and efficiency. While it has proven itself to be useful as an assistive tool, experts recently sought to determine how it could be most beneficial to readers. To do this, researchers deployed AI in multiple scenarios to assess how it might impact radiologists’ performance.
“Artificial intelligence tools for evaluating low-dose CT lung cancer screening examinations are used predominantly for assisting radiologists' interpretations. Alternate utilization scenarios (e.g., use of AI as a prescreener or backup) warrant consideration,” Eui Jin Hwang, MD, PhD, with the Department of Radiology at Seoul National University Hospital in the Republic of Korea, and colleagues wrote in the American Journal of Roentgenology.
The team compared three different utilization scenarios—as an assistant (radiologists interpret all examinations with AI assistance), a prescreener (radiologists only interpret examinations flagged by AI as positive) and a backup reader (radiologists read exams when AI suggests a finding has been missed). Using metrics of time, sensitivity, specificity and recall rates, radiologists’ performance with AI in each respective role was analyzed during their interpretations of nearly 400 lung cancer screening CTs.
Routine AI assistance yielded a higher recall rate (30.3%), lower per-examination specificity (81.1%) but did not significantly impact per-nodule sensitivity compared to radiologists reading solo. When utilized as a prescreening tool, AI assistance reduced recall rates (20.8%) and interpretation times, while also improving per-exam sensitivity. As a backup tool, AI increased recall rates, sensitivity and interpretation times, but decreased specificity.
Though there are benefits in each scenario, the team determined that AI would best serve radiologists when used to prescreen CT scans for signs of lung cancer.
“Among scenarios, only AI as a prescreener demonstrated higher net benefit than interpretation without AI; AI as an assistant had least net benefit,” the group noted. “An approach whereby radiologists only interpret LDCT examinations with a positive AI result can reduce radiologists' workload while preserving sensitivity.”
Learn more about the findings here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.