Predicting Pain: Electroencephalography Signatures of Neural Integration during Experimental Tonic Thermal Pain

Background The identification of reliable neural signatures for pain remains a critical challenge in both clinical and experimental settings. While electroencephalography (EEG) provides a promising avenue for pain assessment, it remains unclear whether phase- or power-based neural integration drives pain-state discrimination. This study investigates functional connectivity and cross-frequency coupling (CFC) as candidate signatures of neural integration for pain prediction. Methods We recorded 62-channel EEG data from 36 healthy participants across experimental conditions, including tonic thermal pain, non-painful warm stimulation and resting states. Functional connectivity within the alpha band and CFC across delta, theta, alpha and low-beta bands was computed using phase- and power-based measures. Machine learning models were trained to classify pain from non-painful conditions, with prediction accuracy serving as an index of neural integration performance. Results Phase-based features outperformed power-based features in tonic thermal pain prediction, which indicated a dominant role for phase synchrony. The strongest topographical signatures are the connectivity involving frontal and occipital regions driven by enhanced alpha-phase connectivity or theta-alpha/delta-theta cross-frequency coupling. Conclusions Our findings demonstrate that phase-based neural integration outperforms power-based integration in characterising tonic pain, and that the inclusion of amplitude information actively reduces discriminability. Phase-based measures of functional connectivity in the alpha band and cross-frequency coupling with other low-frequency oscillations may serve as candidate EEG signatures of pain, offering a data-driven framework with potential applications in research and clinical contexts. Significance Statement This study introduces an EEG-based framework for tonic pain prediction that integrates multiple neural signatures of integration. High classification accuracy between painful and non-painful states provides new machine learning–driven insight into phase-based neural integration, reflected in functional connectivity and CFC. The findings broaden theoretical perspectives on pain processing within machine learning approaches and underscore clinically relevant potential for developing reliable, noninvasive tools to improve pain assessment.