Eye-tracking technology is providing valuable insight into radiologist reading methods of lung nodules on CT images, including the fact that the majority of radiologists visually inspect just a small portion of the lung, according to a study published in the January edition of Radiology.
Detecting pulmonary nodules is a vital component of reading lung images, as both primary lung cancer and metastases typically initially show as pulmonary nodules and variability in detection of them has been reported among radiologists who read CT images, according to the authors Geoffrey D. Rubin, MD, of the Duke Clinical Research Institute, and colleagues.
Studying nodule detection in radiographic chest images with gaze-tracking technology has led to the hypothesis that there are four components to the detection process: orientation, search, recognition and decision making. It is assumed that errors do not occur in the orientation phase—therefore they are relegated to the remaining three (search, recognition and decision making).
The researchers used gaze-tracking technology to study radiologists reading chest CT images to determine variation between readers in the search and recognition phases.
For the study, a remote eye-tracker recorded radiologists interpreting 40 lung CT images with an average of 3.9 synthetic nodules. The researchers defined that radiologists’ gaze volumes (GVs) as the portion of the lung parenchyma within 50 pixels of all gaze points.
The results showed that the detected nodules were within 50 pixels of the nearest gaze point in 990 of 992 correct detections. Radiologists searched 26.7 percent of the lung parenchyma within 3 minutes and 16 seconds, encompassing between 86 and 143 of 157 nodules.
“Our experiment shows that when the center of a radiologist’s gaze is never closer than 50 pixels from a lung nodule, there is less than 1 percent likelihood that the nodule will be detected,” Rubin and colleagues wrote. “The definition of search volume on the basis of this construct results in the observation that, on average, radiologists stopped searching the lungs after only 26.7 percent of the lung was searched.”
The authors noted this fact was interesting, as the radiologists had been instructed to search the entire lung to locate nodules.
“Although the observers searched an average of 26.7 percent of the lung volume, those search volumes encompassed an average of 75 percent of the nodules,” they wrote. “This apparently paradoxical result might suggest that our definition of the search volume might be overly restrictive and that the gaze cone extends beyond 50 pixels from the gaze point; however the recognition and acceptance of a nodule beyond 50 pixels was extremely rare.”
Rubin and colleagues concluded that as momentum builds for widespread lung cancer screening using CT, understanding the variations and limitations of radiologists’ search and detection abilities will be important factors in ensuring effective screening.
“Our results represent a new facet in our understanding of the interpretive process of CT images through the use of gaze tracking; however, further study of the relationship between search and CT interpretation is needed prior to deriving recommendations for improving performance,” they wrote.