Author: Samuel L. Brady, Joseph G. Meier, Elanchezhian Somasundaram, Andrew T Trout 👨🔬
Affiliation: Cincinnati Children's Hospital Medical Ctr, Cincinnati Childrens Hospital Med Ctr 🌍
Purpose:
To analyze eye-tracker gaze data for insights into radiologists’ areas of focus in an observer study of whole-body PET images.
Methods:
For a comparative study of dose matched PET images obtained using two different scanner configurations (5 vs. 6-ring), a single radiologist, blinded to scanner configuration, reviewed 17 pairs of PET images, recording their preference (preferred, slightly preferred, no preference). Paired images were displayed in a 2×2 matrix, with axial and coronal reconstructions positioned in the upper and lower frames. The images were randomly assigned to the left or right side of the frame. A non-intrusive eye tracker was calibrated to record the radiologist's gaze coordinates. Four areas of interest corresponding to each image subsection were defined to calculate total gaze times for every paired image, individual times on coronal and axial images, and their correlation with preference scores were analyzed.
Results:
As previously reported, the radiologist preferred images obtained with the 6-ring detector in 15 out of 17 cases (11 preferred, 4 slightly preferred). The total gaze time for preferred 6-ring images was significantly lower (9.3 seconds) compared to slightly preferred 6-ring (23.1 seconds) and preferred 5-ring images (20.3 seconds) (p = 0.0016). Gaze time between images obtained with the 5-ring and the 6-ring detector showed no significant difference (p = 0.34). For image pairs where the 6-ring images were preferred, the mean gaze time was split between coronal and axial views at 86% to 14%. In contrast, for preferred 5-ring, the split was 70% to 30%, and for slightly preferred 6-ring, it was 60% to 40%.
Conclusion:
Radiologist gaze data on PET images reveals shorter evaluation times for clearly preferred cases, distinct coronal and axial view patterns, and no detector bias. Quantitative eye-tracking data can complement and reveal new insights in human observer studies with subjective assessments.