Cerebral Lateralization Assessment: An Explainable Deep Learning Approach With Channel Attention Mechanism

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging and Signal Processing

Background:

• Cross-frequency coupling (CFC) is crucial for understanding cognitive processes and neural communication.
• Electroencephalography (EEG) signals combined with CFC can detect neurological conditions linked to atypical cerebral lateralization.
• Deep learning (DL) offers advantages for EEG analysis but faces challenges like limited data and high dimensionality.

Purpose of the Study:

• To propose a novel deep learning approach for identifying brain lateralization patterns using inter-hemispheric functional differences derived from CFC.
• To address challenges in DL for EEG analysis, including data scarcity and noise, by employing synthetic data pre-training and a symmetric architecture.
• To enhance the detection of subtle hemispheric differences and provide explainability for clinical applications through a custom attention layer.

Main Methods:

• Development of a novel deep learning model integrating synthetic signal pre-training for phase-amplitude coupling (PAC) analysis.
• Implementation of a symmetric neural network architecture to evaluate inter-hemispheric functional differences.
• Incorporation of a custom attention layer to identify and weigh the importance of relevant EEG channel information.

Main Results:

• The proposed model achieved good classification performance (Area Under the Curve up to 0.85) in assessing brain lateralization.
• The model provided explainable insights into the neural mechanisms underlying the assessed conditions.
• Demonstrated effectiveness in detecting subtle hemispheric differences crucial for clinical applications.

Conclusions:

• The novel deep learning approach effectively reveals lateralization patterns from EEG signals using CFC, outperforming traditional methods.
• The use of synthetic data and attention mechanisms successfully mitigates common challenges in DL-based EEG analysis.
• This method holds promise for early detection of neurological disorders and offers a deeper understanding of their neural basis.