Towards Automated Eye Movement Characterization for Stroke Patients Using Synthetic Video Data and Machine Learning
Hoor Jalo1, Eddie Ström1, Samuel Ollila1
1Chalmers University of Technology, Gothenburg, Sweden.
Studies in Health Technology and Informatics
|August 8, 2025
Summary
Synthetic data can train machine learning models for stroke detection. This approach aids in developing automated systems for prehospital stroke assessment by analyzing characteristic eye movements.
Area of Science:
- Medical technology
- Artificial intelligence in healthcare
- Neurology
Background:
- Stroke is a leading cause of disability and death, necessitating rapid prehospital diagnosis.
- Effective prehospital stroke assessment is hindered by a lack of diverse patient video data for machine learning model development.
Purpose of the Study:
- To investigate the utility of synthetic video data for training machine learning models for stroke detection.
- To develop and evaluate machine learning models for automated assessment of stroke-related eye movements in prehospital settings.
Main Methods:
- Generation of 73 synthetic videos using 3D modeling and animation to simulate characteristic stroke patient eye movements.
- Development and comparison of four distinct machine learning models, including Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU).
Main Results:
- The Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) models demonstrated superior performance.
- Achieved over 84% accuracy, precision, sensitivity, specificity, and F1-Score in stroke detection using synthetic data.
Conclusions:
- Synthetic data holds significant promise for developing robust machine learning models in healthcare applications.
- ML-driven video analysis of eye movements can support automated prehospital stroke assessment, improving patient care.
