Lightweight Design for EEG Acquisition: a Trade-off between Reconstruction Accuracy and Data Efficiency

Electroencephalography (EEG) is a non-invasive method for measuring and recording electrical activity in the brain, which is useful in identifying neurological illnesses. Traditional EEG configurations involve a network of electrodes inserted on the scalp. They are usually limited to short-term recordings in monitored healthcare settings. However, the trend toward personalized treatment and remote healthcare has resulted in a higher demand for continuous long-term EEG monitoring in real world settings such as houses. Long-term monitoring captures brain activity during daily activities, providing useful information to control chronic neurological diseases. A key challenge in remote EEG monitoring lies in achieving a balance between reducing the number of electrodes for practicality and maintaining the data quality required for clinical diagnosis. This project evaluated three methods, Linear Regression (LR), Principal Component Analysis (PCA), and Multi-Layer Perceptrons (MLPs), for reconstructing EEG signals from a reduced number of electrodes. The results showed that MLP outperformed others in the testing, achieving an average R2 score of 0.5452. In addition, electrodes for the central and parietal channels provided the most accurate reconstructions in all methods, emphasizing their critical role in achieving robust signal quality and electrode placement around the scalp. A composite performance metric was introduced to evaluate the trade-off between reconstruction accuracy and data efficiency, underscoring the importance of balancing model complexity with practical considerations for telemedicine applications. These findings highlight the feasibility of achieving reliable EEG reconstructions with fewer electrodes, paving the way for lightweight, energy-efficient systems for remote health monitoring.