Detecting acute pain signals from human EEG

Advances in human neuroimaging has enabled us to study functional connections among various brain regions in pain states. Despite a wealth of studies at high anatomic resolution, the exact neural signals for the timing of pain remain little known. Identifying the onset of pain signals from distributed cortical circuits may reveal the temporal dynamics of pain responses and subsequently provide important feedback for closed-loop neuromodulation for pain.

Here we developed an unsupervised learning method for sequential detection of acute pain signals based on multichannel human EEG recordings. Following EEG source localization, we used a state-space model (SSM) to detect the onset of acute pain signals based on the localized regions of interest (ROIs).

We validated the SSM-based detection strategy using two human EEG datasets, including one public EEG recordings of 50 subjects. We found that the detection accuracy varied across tested subjects and detection methods. We also demonstrated the feasibility for cross-subject and cross-modality prediction of detecting the acute pain signals.

In contrast to the batch supervised learning analysis based on a support vector machine (SVM) classifier, the unsupervised learning method requires fewer number of training trials in the online experiment, and shows comparable or improved performance than the supervised method.

Our unsupervised SSM-based method combined with EEG source localization showed robust performance in detecting the onset of acute pain signals.