Prediction of Radiological Diagnostic Errors from Eye Tracking Data Using Graph Neural Networks and Gaze-Guided Transformers

Predicting radiological diagnostic errors is crucial for improving the effectiveness of patient care. This work analyzes longitudinal gaze paths of radiologists to find patterns associated with diagnostic errors, combining target medical images, gaze fixation data, and accumulated gaze maps. A graph neural network and gaze-guided transformer are trained to detect fixation patterns predictive of errors on a large set of chest radiograph readings, achieving favorable accuracy and ROC AUC results. :contentReference[oaicite:0]{index=0}

Graphs in Biomedical Image Analysis