An EEG-Based Framework for Sleep Quality Assessment and Modulation with Conditional Convolutional Diffusion Modeling

Sleep quality assessment and modulation are essential for sleep health monitoring. However, multi-channel electroencephalography (EEG) systems are unsuitable for long-term and daily applications, while EEG-based sleep quality classification is often limited by data scarcity and class imbalance. To address these challenges, an integrated framework for sleep quality classification and modulation is proposed. First, a wearable EEG acquisition device is developed, and its effectiveness is validated through comparative analysis with a standard multi-channel EEG system, demonstrating high signal consistency through both signal-level and agreement-level analyses. Second, a transcutaneous electrical nerve stimulation approach targeting the greater occipital nerve (GON-TENS) is proposed for sleep modulation, with experimental results indicating improved sleep structure and overall sleep quality. Third, a conditional convolutional diffusion model (CCDM) is proposed for EEG data augmentation to alleviate class imbalance, particularly between high- and low-quality sleep categories. Quantitative evaluations using root mean squared error, correlation coefficient, and SNR confirm the effectiveness of CCDM in generating high-fidelity EEG signals. Moreover, within a CNN-based sleep quality classification model, the proposed CCDM consistently improves overall classification performance compared with existing augmentation methods, achieving a macro-averaged F1-score of 82.2%, an accuracy of 88.6%, and a Cohen's kappa of 0.81. These results demonstrate the potential of the proposed framework, offering an effective solution for sleep quality assessment and modulation in practical applications.