By Jennie McKee
The adoption of a lung nodule detection system into the clinical workflow of a community practice significantly reduced reading times, according to a study presented Tuesday. The AI tool also improved radiologists’ confidence regarding reporting lung nodules.
“The assumption that AI-assisted radiology is inherently superior to traditional methods is not always accurate,” said Baskaran Sundaram, MD, professor of radiology at Thomas Jefferson University in Philadelphia.
“It is essential to assess whether AI tools enhance accuracy, productivity or efficiency,” Dr. Sundaram said. “Moreover, it's crucial to consider if these tools exacerbate or mitigate burnout.”
As early adopters of AI tools, Dr. Sundaram and colleagues are actively exploring strategies to justify the purchase, retention and replacement of these technologies. “This case study is a prime example of this ongoing effort,” he said.
The pilot study assessed the interpretation times of nine community-based radiologists with and without a lung nodule detection algorithm after integrating it into the institution’s PACS workflow. The tool can automatically detect lung nodules on chest CT and report the nodules’ characteristics, such as type and diameter.
Throughout a 90-day period, the investigators recorded the reading time for all chest CTs interpreted with and without the algorithm.
During the study period, the radiologists read a total of 738 scans aided by the detection algorithm, and 1,192 scans without it. The investigators found that using the AI software resulted in a statistically significant reduction in reading times. Without AI, the average reading time was 13±11.4 minutes; however, with AI, the average reading time was 12.2±10.0 minutes.
In addition, anonymous survey responses from the radiologists who participated in the study indicated that using the AI system increased their confidence in reporting lung nodules using the AI system.
Meticulous planning was crucial for obtaining reliable measurements, Dr. Sundaram noted.
“By selecting a single, non-teaching division with relatively uniform radiologist roles and responsibilities, we established a controlled environment,” he said. “Integrating the tool seamlessly into the PACS workflow and providing comprehensive training to radiologists ensured familiarity and ease of use. These strategic measures were pivotal in ensuring the credibility of our results.”
Dr. Sundaram noted that although the study results showed a benefit to using a lung-nodule detection algorithm when reading chest CTs, this analysis presents the findings of a single pilot study evaluating a specific AI tool within a community practice over a controlled period.
“It is important to note that these results do not warrant the widespread adoption of any AI system,” he said. “As demonstrated in the literature, outcomes can vary significantly across different settings and implementations.”
Assessing the radiologists’ accuracy was outside the purview of this study; however, Dr. Sundaram noted that conducting such a study using a suitable definition and measurement parameters could offer valuable insights.
Access the presentation, “Workflow Efficiency Benefits of a Lung Nodule Detection AI Program,” (T1-SSCH05-4) on demand at RSNA.org/MeetingCentral.
© 2024 RSNA.
The RSNA 2024 Daily Bulletin is the official publication of the 110th Scientific Assembly and Annual Meeting of the Radiological Society of North America. Published online Sunday, December 1 — Friday, December 6.
The RSNA 2024 Daily Bulletin is owned and published by the Radiological Society of North America, Inc., 820 Jorie Blvd., Suite 200, Oak Brook, IL 60523.
