Remembered or Forgotten?-An EEG-Based Computational Prediction Approach.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Predicting memory performance (remembered vs. forgotten) has significant applications in learning and disease diagnosis.
• Subsequent Memory Effects (SMEs) reveal statistical differences in electroencephalography (EEG) signals related to memory outcomes.
• EEG signals contain information predictive of future memory performance.

Purpose of the Study:

• To propose a computational approach for predicting memory performance using EEG signals based on SMEs.
• To develop and validate a convolutional neural network (CNN) model for this prediction task.

Main Methods:

• Devised a CNN for EEG, named ConvEEGNN, integrating feature extraction and classification stages.
• Collected EEG data during an auditory memory task using scalp electrodes.
• Utilized both pre-stimulus and during-stimulus EEG periods for prediction.

Main Results:

• The ConvEEGNN achieved an average prediction accuracy of 72.07%, outperforming other methods.
• Both pre-stimulus and during-stimulus EEG periods contributed comparably to memory performance prediction.
• Network connection weights identified prominent EEG channels, aligning with previous SME research.

Conclusions:

• The proposed ConvEEGNN model effectively predicts memory performance from EEG signals.
• EEG signals, particularly from pre-stimulus and during-stimulus periods, hold valuable predictive information for memory.
• This computational approach offers a promising tool for understanding and potentially enhancing memory processes.