Predicting Performance in Working Memory During the Waking Period by Applying a Convolutional Neural Network to EEG...

Area of Science:

• Neuroscience
• Chronobiology
• Artificial Intelligence

Background:

• Circadian rhythms significantly influence memory performance.
• Electroencephalograms (EEG) capture brain activity relevant to memory and circadian patterns.
• EEG offers potential for detecting diurnal variations in memory function.

Purpose of the Study:

• To investigate the efficacy of convolutional neural networks (CNNs) in predicting memory task performance based on EEG signals.
• To assess the predictive power of EEG relative power and raw waveform data for diurnal memory changes.
• To explore the impact of personalized data on CNN model performance for chronotype-based memory prediction.

Main Methods:

• Recorded EEG signals from participants performing N-back tasks at morning (8-9 a.m.) and afternoon (3-4 p.m.) sessions.
• Trained CNN models using both relative power and raw waveform EEG data.
• Evaluated CNN model accuracy in predicting task times and the effect of participant-specific training data.

Main Results:

• No significant difference in memory task performance was observed between morning and afternoon sessions.
• Significant differences in EEG relative power were detected between the two time points.
• The CNN model utilizing relative power data achieved higher accuracy in predicting task times compared to the raw waveform model.
• Model performance decreased when tested on data from participants not included in the training set.

Conclusions:

• EEG signals, particularly relative power, hold significant predictive potential for diurnal memory variations when analyzed with CNNs.
• The effectiveness of EEG-based memory prediction is enhanced by personalized models that account for individual chronotypes.
• Future applications may benefit from tailored classification systems for practical chronotype-specific memory enhancement strategies.