Objective: To investigate machine learning for linking image content, human perception, cognition, and error in the diagnostic interpretation of mammograms.
Methods: Gaze data and diagnostic decisions were collected from six radiologists who reviewed 20 screening mammograms while wearing a head-mounted eye-tracker. Texture analysis was performed in mammographic regions that attracted radiologists’ attention and in all abnormal regions. Machine learning algorithms were investigated to develop predictive models that link: (i) image content with gaze, (ii) image content and gaze with cognition, and (iii) image content, gaze, and cognition with diagnostic error. Both group-based and individualized models were explored.
Results: By pooling the data from all radiologists machine learning produced highly accurate predictive models linking image content, gaze, cognition, and error. Merging radiologists’ gaze metrics and cognitive opinions with computer-extracted image features identified 59% of the radiologists’ diagnostic errors while confirming 96.2% of their correct diagnoses. The radiologists’ individual errors could be adequately predicted by modeling the behavior of their peers. However, personalized tuning appears to be beneficial in many cases to capture more accurately individual behavior.
Conclusions: Machine learning algorithms combining image features with radiologists’ gaze data and diagnostic decisions can be effectively developed to recognize cognitive and perceptual errors associated with the diagnostic interpretation of mammograms.