Discrimination of Radiologists' Experience Level Using Eye-Tracking Technology and Machine Learning: Case Study.

Area of Science:

• Medical imaging analysis
• Radiology informatics
• Human-computer interaction

Background:

• Perception-related errors are a significant cause of diagnostic mistakes in radiology.
• Radiologists employ visual search patterns, but qualitative descriptions are unreliable, hindering quality improvement.
• Discrepancies between reported and actual visual search patterns impact patient care.

Purpose of the Study:

• To develop an objective method for differentiating radiologists using eye-tracking data.
• To discriminate between radiologists based on subconscious visual inspection behavior.
• To leverage raw gaze or fixation data for expertise assessment.

Main Methods:

• A novel discretized feature encoding based on spatiotemporal binning of fixation data was developed.
• Machine learning classifiers used encoded eye-movement data to differentiate faculty and trainee radiologists.
• Performance was evaluated using AUC, accuracy, F1-score, sensitivity, and specificity, compared to state-of-the-art methods.

Main Results:

• The proposed feature encoding method outperformed current state-of-the-art techniques in differentiating radiologists by experience.
• An average performance gain of 6.9% was observed compared to traditional features.
• Significant accuracy improvements were noted across different eye-tracker datasets (Tobii: 6.41%, EyeLink: 7.29%).

Conclusions:

• The spatiotemporal discretization approach offers an effective, objective method for evaluating radiologists' expertise.
• Validated across diverse datasets, the method can inform targeted interventions and training strategies.
• This research provides reliable assessment tools to address perception-related errors and enhance patient care in radiology.