INSIGHT: Combining Fixation Visualisations and Residual Neural Networks for Dyslexia Classification From Eye-Tracking...

Area of Science:

• Neuroscience
• Computer Science
• Developmental Psychology

Background:

• Traditional dyslexia diagnosis relies on paper-and-pencil tests.
• Technological advancements like eye-tracking and AI offer improved diagnostic potential.
• A gap exists between scientific understanding and practical diagnostic tools for dyslexia.

Purpose of the Study:

• To propose and evaluate INSIGHT, a novel method for dyslexia detection.
• To integrate eye-tracking visualization with artificial intelligence for enhanced diagnostics.
• To provide a robust framework for early and accurate dyslexia diagnosis.

Main Methods:

• Developed INSIGHT, combining a visualization phase (Fix-images from eye-tracking data) and a classification phase (ResNet18 neural network).
• Collected eye-tracking fixation data from 35 child participants (13 with dyslexia, 22 controls) during reading tasks.
• Cross-tested the method on an independent dataset of Danish readers to assess generalizability.

Main Results:

• The INSIGHT method achieved a highest accuracy of 86.65% in detecting dyslexia in the initial cohort.
• Cross-testing on an independent dataset yielded a notable accuracy of 86.11%.
• Fix-images effectively visualized reading difficulties, and the neural network accurately classified them.

Conclusions:

• The INSIGHT method offers a robust and generalizable framework for early dyslexia detection.
• This AI-driven approach provides detailed insights into reading-related eye movement patterns.
• The findings support the potential for more personalized and effective dyslexia interventions.