Decoding food odor-evoked EEG signals: odor recognition and brain region analysis using Mixed Feature Attention Network (MFANet)

Olfactory cues play a crucial role in food selection, with undesirable odors often being perceived as indicative of spoiled or inferior food. However, odor perception is highly subjective and prone to low reproducibility. Electroencephalographic (EEG) signals induced by odors contain rich information that can be leveraged to decode food odors. In this study, we employed EEG and source localization signals triggered by 8 different food odors as input data and proposed a Mixed Feature Attention Network (MFANet) to differentiate between food odors. Firstly, we established an experimental paradigm for olfactory EEG data collection and source localization signal processing. Secondly, we introduced a mixed feature data mining strategy, which effectively integrates both global and local features of the data. Experimental results demonstrated that MFANet achieved an average accuracy of 97.35% in distinguishing between the 8 food odors. Moreover, using standardized low resolution brain electromagnetic tomography (sLORETA) source localization, we identified significant differences in brain regions activated by different food odors, particularly in the right temporal lobe, where activation differences between pleasant and unpleasant odors were statistically significant (t > 5, P < 0.001). In conclusion, MFANet effectively mined the information contained in olfactory EEG data and successfully differentiated between the eight food odors. It also captured subtle differences in brain activation patterns induced by food odors. These findings suggest that MFANet holds potential for applications in the diagnosis and treatment of olfactory dysfunction.