Machine learning evaluates changes in functional connectivity under a prolonged cognitive load

Area of Science:

• Neuroscience
• Machine Learning
• Cognitive Science

Background:

• Machine learning models excel at neuroimaging analysis but often function as black boxes.
• Understanding the internal algorithms and input features is crucial for interpreting model decisions.
• High-dimensional neuroimaging data presents challenges for model interpretability.

Purpose of the Study:

• To classify cognitive brain states using functional connectivity data.
• To address the black-box problem by selecting and interpreting relevant input features.
• To investigate alterations in cortical synchrony under prolonged cognitive load.

Main Methods:

• Applied machine learning to high-dimensional functional connectivity data.
• Focused on feature selection and interpretation to enhance model transparency.
• Analyzed changes in cortical synchrony during cognitive tasks.

Main Results:

• Developed a robust machine learning model for classifying cognitive brain states.
• Identified key input features driving the classification performance.
• Revealed percept-related prestimulus connectivity changes.

Conclusions:

• Machine learning can effectively classify cognitive brain states from neuroimaging data.
• Feature selection and interpretation are vital for understanding black-box models in neuroscience.
• The study offers a more insightful approach than traditional trial-averaged statistical analysis.