Related Experiment Video
Updated: Sep 30, 2025
13:40
Combining Computer Game-Based Behavioural Experiments With High-Density EEG and Infrared Gaze Tracking
Published on: December 16, 2010
16.8K
Gaming behavior and brain activation using functional near-infrared spectroscopy, Iowa gambling task, and machine
Denis Kornev1, Stanley Nwoji1, Roozbeh Sadeghian2
1Information System Engineering and Management Program, Harrisburg University of Science and Technology, Harrisburg, Pennsylvania, USA.
Brain and Behavior
|March 15, 2022
Summary
Machine learning algorithms effectively analyze brain activity during gaming, correlating functional near-infrared spectroscopy signals with task performance. This approach accurately predicts cognitive states and decision-making processes in participants.
Area of Science:
- Cognitive Neuroscience
- Neuroimaging
- Machine Learning Applications
Background:
- Investigating the relationship between brain activation and cognitive tasks is crucial for understanding decision-making.
- Functional near-infrared spectroscopy (fNIRS) offers a non-invasive method to measure brain hemodynamics.
- Machine learning (ML) presents a powerful tool for analyzing complex neuroimaging data.
Purpose of the Study:
- To evaluate the performance of ML algorithms in correlating human brain activation during gaming with numerical parameters.
- To test the hypothesis that an integrated feature extraction platform can distinguish psychosomatic conditions during gaming using fNIRS.
- To assess the predictive accuracy of various ML models for cognitive task performance.
Main Methods:
- Combined the Iowa Gaming Task (IGT) simulator with fNIRS neuroimaging.
- Extracted, averaged, and synchronized fNIRS features with IGT data.
- Employed ML algorithms (multiple regression, trees, SVM, ANN, random forest) trained and validated using cross-validation, with R-squared and RMSE as performance metrics.
Main Results:
- Highest correlation between oxy-hemoglobin signal and IGT score observed in block 4 (0.57 for signal feature).
- ML algorithms showed acceptable performance, with RMSE increasing and R-squared decreasing across blocks, reflecting a transition from uncertainty to certainty.
- Support Vector Machine with a radial basis function kernel consistently yielded the highest prediction accuracy (lowest RMSE) across most IGT blocks.
Conclusions:
- ML models are applicable and capable tools for evaluating cognitive neuroimaging task results.
- Hemodynamic responses, as measured by fNIRS, react to accelerated decision-making processes, indicating increased significance.
- The study demonstrates the potential of ML in neuroimaging for understanding cognitive processes during interactive tasks.
