When Seeing Is Not Believing: Eye-Tracking Evidence of Missed Prostate Cancer on Mpmri
Abstract
Purpose
Accurately identifying clinically significant prostate cancer (PCa) lesions remains a difficult task. Although tumor characteristics associated with false-negative lesions have been investigated, the potential radiologist search strategy issues underlying missed lesions are not understood. This study aimed to characterize search strategy features associated with detected and missed PCa using eye-tracking technology.
Methods
Eye tracking data were collected from four radiologists with varying levels of experience during multiparametric MRI (mpMRI) assessment of 40 patients. Radiologist contours and eye movements were spatially aligned with co-registered PCa contours derived from whole-mount histopathology. Search strategy metrics were calculated and included lesion foveation incidence, cumulative foveation duration, and average foveation duration. Search strategy metrics were compared between true-positive and false-negative lesions. Recall and precision quantitative metrics were calculated before and after exclusion of equivocal contours and stratified by radiologist experience and cancer grade. Recall measures the fraction of true-positive lesions correctly identified and precision measures the fraction of radiologist contours that correspond to true-positives.
Results
All true-positive lesions were foveated upon, whereas 2-4% of false-negative lesions were never foveated upon. True-positive lesions exhibited 2-4 times longer cumulative foveation duration than false-negative lesions. The average fixation duration on true-positive lesions was greater on at least one mpMRI sequence compared to an average fixation duration on false-negative lesions. The least experienced radiologist demonstrated lower recall for low-grade and small lesions, while recall for higher-grade cancers was similar across experience levels. Excluding equivocal contours increased precision for all radiologists, but reduced recall for radiologists with less experience.
Conclusion
Missed prostate cancers on mpMRI are associated with incomplete or insufficient search and reduced diagnostic confidence. By revealing perceptual and attentional contributors to false-negative lesions, this study provides insight into PCa detection and establishes a foundation for perceptually informed training, workflow optimization, and attention-aware diagnostic support tools.