Mapping Scalp-Level EEG Representations from a Multimodal In-Ear Device for Mental Fatigue Assessment: A Stacked LSTM-Based Approach

Abstract

Construction activities impose substantial cognitive demands on workers, making mental fatigue assessment critical for occupational health and safety (OHS) management. Portable wearable electroencephalography (EEG) devices offer practical advantages in mental fatigue identification. However, they are typically limited to a few channels, restricting the spatial coverage of brain activity and potentially limiting the ability to capture fatigue-related spatial patterns. To address this limitation, a sensor-level scalp EEG representation mapping framework was developed to enhance spatial information in fatigue assessment. First, a stacked long short-term memory (LSTM)-based regression model is developed to approximate multichannel scalp EEG representations from preprocessed single-channel in-ear EEG and electrocardiography (ECG) signals. Quantitative evaluation demonstrates stable predictive performance, with mean absolute error ( MAE ) and root mean square error ( RMSE ) both below 24    μV across fatigue states. Second, the predicted scalp EEG signals are transformed into EEG topographic maps to enhance spatial interpretability. The proposed framework enables portable yet spatially informative fatigue assessment and provides methodological support for future OHS-oriented monitoring systems.