Loading...

Eye Tracking using EEG Signals During Simultaneous Functional Magnetic Resonance Imaging

Mortazavi, Abolfazl | 2026

81 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58845 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Ghazizadeh Ehsaei Ehsaei, Ali; Shamsollahi, Mohammad Bagher
  7. Abstract:
  8. In certain behavioral neuroscience studies, precise monitoring of the subject's gaze direction and duration is of particular importance. However, the use of conventional eye-tracking systems in functional Magnetic Resonance Imaging (fMRI) environments faces significant challenges due to strong magnetic fields, physical constraints, and high costs. Aiming to provide a cost-effective alternative method, this study investigates the feasibility of utilizing Electroencephalogram (EEG) signals recorded simultaneously with fMRI to track eye direction and movement. To achieve this objective, several visual tasks were designed, including five-class, three-class, and passive viewing tasks. EEG and eye-tracking data were collected from participants under two conditions: outside the scanner and inside the MRI scanner. In the pre-processing stage, while gradient and pulse artifacts were removed (for in-scanner data), the removal of independent components (ICA) was omitted to preserve the information embedded in ocular artifacts. Time and frequency features (such as Hjorth parameters and Alpha/Beta band powers) were extracted and employed to train machine learning models, including Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest, and XGBoost. The results indicated that in the out-of-scanner environment, the SVM model with a linear kernel yielded the best performance, achieving an average accuracy of 99% in the passive viewing task and 94% in the three-class task. An accuracy of approximately 86% was also obtained in the five-class tasks. In the in-scanner MRI experiments, despite the degradation in signal quality, the models demonstrated acceptable performance; the XGBoost model achieved the best result in the in-scanner three-class task with 85% accuracy, while the SVM model recorded 100% accuracy in the passive viewing task. Furthermore, the results demonstrated that employing Independent Component Analysis (ICA) led to a 10% reduction in model performance, and transfer learning between subjects and different tasks was associated with a significant decrease in accuracy. This study suggests that EEG signals contain rich information regarding eye movements and can be utilized as an efficient tool for tracking gaze direction in complex environments such as fMRI, without the need for additional hardware
  9. Keywords:
  10. Eye Tracking ; Electroencphalogram Signal ; Functional Magnetic Resonance Imaging (FMRI) ; Machine Learning ; Signal Registeration ; Simultaneous Electroencphalogram (EEG)-Functional Magnetic Resonance Imaging (FMRI)Signals

 Digital Object List

 Bookmark