Deep Convolutional Neural Networks and Transfer Learning for Measuring Cognitive Impairment Using Eye-Tracking in a...

Area of Science:

• Neuroscience
• Medical Technology
• Ophthalmology

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by memory loss, amyloid plaques, and neurofibrillary tangles.
• Mild cognitive impairment (MCI) represents an early symptomatic stage of AD, offering a critical window for intervention before dementia onset.
• Early detection of cognitive impairment is crucial for managing Alzheimer's disease and improving patient outcomes.

Purpose of the Study:

• To develop a mobile version of the Visuospatial Memory Eye-Tracking Test (VisMET) for widespread and efficient administration.
• To adapt the VisMET for use on tablet devices, enabling broader accessibility for cognitive assessment.
• To evaluate the efficacy of the mobile VisMET in identifying cognitive impairment associated with Alzheimer's disease.

Main Methods:

• The VisMET was implemented on iPad devices, utilizing a deep neural network trained with transfer learning to analyze eye gaze patterns.
• Eye movement data from 250 individuals were collected and processed to extract memory-related features for cognitive status assessment.
• A minimal eye-tracking calibration error of 2 cm was enforced to optimize accuracy.

Main Results:

• The mobile VisMET demonstrated the ability to identify mild to severe cognitive impairment with an initial accuracy of 70%.
• Enforcing a calibration error threshold of 2 cm improved the test's accuracy to 76%.
• This accuracy level is comparable to that achieved with commercial eye-tracking hardware.

Conclusions:

• The developed mobile VisMET effectively estimates the presence of cognitive impairment, offering a viable screening tool.
• The study validates a portable and user-friendly method for assessing cognitive function related to Alzheimer's disease.
• The widespread availability of tablet devices positions this mobile VisMET for global scalability in cognitive health monitoring.