Although computer-aided detection (CAD) technology has been shown to be beneficial in locating lung nodules on chest CT that would otherwise go undetected, inspecting CAD hits in addition to a traditional readout adds to the total time needed to inspect a CT volume, said presenter Martin Tall.
As a result, the study team is investigating methods of diagnostic support in which the visual search pattern of the reader is fed into the system.
"This holds great potential as it opens up a window to the cognitive processes that underlie visual search and bridges the gap between man and machine," Tall said.
The researchers have employed tracking devices to establish a platform for acquiring real-time eye gaze data during inspection of volumetric datasets. This platform can tell where visual attention is directed with a spatial resolution of 5 mm to 10 mm on-screen at a 20-msec temporal resolution, Tall said.
"Using our platform we are investigating the perceptual capabilities of the visual system in relation to medical images," Tall said. "For example, we are studying the extent of the peripheral visual field for detecting nodules in volumetric lung CT data."
Tall said the researchers were able to record volumetric gaze data to build 3D gaze paths through large volumetric datasets, providing insight into what regions were inspected.
"More importantly, these could be used to highlight areas in the dataset that never received attention or detect systematic errors in search to ensure a more complete examination," Tall told AuntMinnie.com. "Furthermore, the approach holds potential when combined with a localized CAD algorithm. The reader would only be notified of a potential lesion if a candidate was found in an unattended area."