Sleep Brain and Cardiac Activity Predict Cognitive Flexibility and Conceptual Reasoning Using Deep Learning.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Sleep Science

Background:

• The link between sleep microstructure and specific cognitive functions is not well understood.
• Executive functions, including cognitive adaptability and conceptual reasoning, are crucial for daily performance.

Purpose of the Study:

• To investigate the predictive power of deep learning models for executive functions using overnight physiological data.
• To introduce and evaluate CogPSGFormer, a novel deep learning architecture for analyzing polysomnographic data.

Main Methods:

• Developed CogPSGFormer, a multi-scale convolutional-transformer model processing multi-modal polysomnographic data (ECG, EEG).
• Integrated EEG power bands and heart rate variability features for comprehensive signal analysis.
• Evaluated the model on 817 individuals from the STAGES dataset using cross-validation.

Main Results:

• CogPSGFormer achieved 80.3% accuracy in classifying cognitive performance (low vs. high) based on Penn Conditional Exclusion Test (PCET) scores.
• Demonstrated the model's effectiveness in predicting cognitive performance from sleep-derived physiological signals.
• Highlighted the utility of multi-scale feature extraction and multi-modal learning.

Conclusions:

• Deep learning models, like CogPSGFormer, can effectively predict executive functions from sleep physiological data.
• The study underscores the potential of leveraging sleep microstructure for understanding and forecasting cognitive performance.
• The developed framework and code are publicly available to promote further research and reproducibility.