Pain Classification using Evoked EEG Induced by Thermal Grill Illusion - Deep Neural Network Approach

As the quantification of pain has emerged in biomedical engineering today, studies have been developing biomarkers associated with pain actively by measuring bio-signals such as electroencephalogram (EEG). Recently, some EEG studies of cold and hot pain have been reported. However, they used one type of stimulus condition for each trial and a relatively long stimulation time to collect EEG features. In this study, EEG signals during Cool (20 °C), Warm (40 °C), and Thermal Grill Illusion (TGI, 20-40 °C) stimuli were collected from 43 subjects, and were classified by a deep convolutional neural network referred to as EEGNet. Three binary classifications for the three conditions (TGI, Cool, Warm) were conducted for each subject individually. Classification accuracies for TGI-Cool, TGI-Warm, and Warm-Cool were 0.74±0.01, 0.71±0.01, and 0.74±0.01, respectively. For subjects who rated the TGI significantly hotter than the Warm stimulus, the classification accuracy for TGI-Cool (0.74±0.01) was significantly higher than for TGI-Warm (0.71±0.01). In contrast, the classification accuracy for TGI-Cool (0.72±0.03) did not differ statistically from TGI-Warm (0.73±0.01) in subjects without illusion. We found that the TGI and Cool stimuli were classified better than the TGI and Warm stimuli, implying that objective EEG features are consistent with subjective behavioral results. Further, we observed that most discriminative features between the TGI and the Cool or Warm conditions appeared in the parietal area for subjects who perceived the illusion. We postulate that the somato-sensory cortex may be activated when TGI is perceived to be hot pain.