🧠 Collection and Visualization of Eye Tracking Data for Machine Learning

Honours Thesis by Ripudaman Singh

Supervised by Dr. Matthew Hamilton

Memorial University of Newfoundland | December 2024

📄 Abstract

Developed a framework for:

• Annotating mammograms automatically in 3D Slicer.
• Collecting radiologist eye-tracking data using Tobii Pro.
• Visualizing gaze data as heatmaps to understand diagnostic focus.
• Creating datasets to support machine learning in breast cancer detection.

👀 Motivation & Introduction

• Challenge: Variability in mammogram diagnosis due to subjectivity and complexity.
• Solution: Use of eye-tracking to capture radiologists’ diagnostic attention.
• Goal: Combine human gaze data with image annotations to enhance AI interpretability.

📚 Literature Review Highlights

• Eye-Tracking Features: Gaze data, pupil size, and blinks improve ML performance.
• Averaging Eye Data: Averaging gaze points from both eyes yields better accuracy unless binocular misalignment exists.