Monday, November 30 | 9:30 a.m.-10:30 a.m. | M3-SSIN02-4 | Room E450B
In this talk, researchers will explain how usage logs from AI-assisted reporting tools can effectively measure cognitive load.
Cognitive load in radiology is hard to measure at scale, noted presenter Steven Rothenberg, MD, of Thomas Jefferson University, and colleagues. Existing approaches rely on subjective surveys, eye-tracking, or rough surrogates such as relative value units (RVUs) and exam counts, which miss case-level demand, according to the researchers.
To explore an alternative, the group retrospectively analyzed 4,102 final reports from a 30-day deployment of a commercial AI-assisted reporting tool used by 12 radiologists at four practices. A large language model extracted case details from each report. After assessing five independent measures of case complexity, the researchers found that tool use climbed with all five complexity measures.
For example, it rose from about 11% to 33% as the number of billed anatomic regions increased. The pattern held for all 12 radiologists, even though they never coordinated, and it was stable across modalities and site, according to the researchers.
“When radiologists choose to invoke AI-assisted reporting, they are signaling which cases they perceive as cognitively demanding,” the authors wrote. “Practices can use their own adoption logs as a workload-monitoring tool, and investigators should report adoption-by-complexity gradients alongside any aggregate finding.”
Hear the full results at this Monday morning presentation.
