Related Experiment Video
Updated: Jan 16, 2026
05:54
Eye-tracking to Distinguish Comprehension-based and Oculomotor-based Regressive Eye Movements During Reading
Published on: October 18, 2018
6.7K
Scanning faces: a deep learning approach to studying eye movements in prosopagnosia
Atlas Kazemian1, Ipek Oruc1, Jason J S Barton1,2,3
1Department of Ophthalmology and Visual Sciences, University of British Columbia, Vancouver, BC, Canada.
Frontiers in Neurology
|September 26, 2025
Summary
Artificial intelligence can identify face scanning differences in acquired prosopagnosia (face blindness) using just a few fixations. Developmental prosopagnosia scanpaths, however, appear as degraded normal patterns, not distinct biases.
Area of Science:
- Neuroscience
- Computer Science
- Ophthalmology
Background:
- Healthy individuals exhibit fixation biases during face scanning, focusing on informative facial regions.
- Individuals with prosopagnosia (face blindness) may display atypical face scanning patterns.
Purpose of the Study:
- To investigate if artificial intelligence (AI) can detect characteristic face scanning markers associated with prosopagnosia.
- To differentiate between acquired and developmental prosopagnosia based on eye movement patterns.
Main Methods:
- Utilized deep learning and image classification to analyze eye fixation data during face recognition tasks.
- Trained convolutional neural networks on visualized scanpaths to distinguish between prosopagnosia subtypes and controls.
- Determined the optimal number of fixations for accurate classification.
Main Results:
- Acquired prosopagnosia was accurately classified from controls using only four fixations (80% AUC), with a tendency to fixate the lower face and right eye.
- Developmental prosopagnosia required 16 fixations for classification (69% AUC) and showed peripheral fixation biases.
- AI models struggled to classify developmental prosopagnosia as distinct from controls, classifying them as similar to normal scanpaths.
Conclusions:
- Deep learning effectively identifies distinct behavioral markers in acquired prosopagnosia through minimal fixations.
- Developmental prosopagnosia scanpaths are characterized by degraded patterns rather than unique biases.
- AI demonstrates potential for identifying complex visual processing disorder markers.
