Monday, November 30 | 9:30 a.m.-10:30 a.m. | M3-SSIN02-3 | Room E450B
AI doesn’t always lighten radiologists’ mental workload, and in some cases adds to it, researchers will report.
AI is often claimed to reduce radiologists’ cognitive burden, but its actual effects on cognitive load aren’t well understood, noted presenter Bonnie Armstrong, PhD, of Stanford University, and colleagues. Drawing on cognitive load theory, the group measured three kinds of demand across four clinically deployed AI tools: those inherent to the task, those created by design and workflow, and the effort that supports learning.
In the prospective single-center study, the researchers observed and interviewed 36 radiologists, both experts and trainees, as they performed four tasks with and without AI. The tasks were pediatric bone age estimation, adult lung air trapping estimation, pediatric lung nodule detection, and adult stroke penumbra estimation. Radiologist ratings were then used to build a reusable codebook for scoring cognitive load.
The effects depended on context, according to the group. When radiologists agreed with the bone age tool’s results, their total cognitive load fell by more than half. The air trapping and stroke penumbra tools, by contrast, increased load even when readers agreed with the results. The lung nodule tool raised load the most, largely because radiologists had to work through many false positives, according to the researchers.
“Evaluators should determine whether AI-related load adds value or reflects unnecessary [extraneous cognitive load] before adoption,” the researchers wrote.
Stop in on this Monday morning session for more information.
