EEG Decoding of Auditory Spatial Attention Based on Visibility Graph and Machine Learning

Peixin Wang, Tasleem Kausar, Heng Li, Siyang Jiang, Yingying Rong, Yun Lu

Abstract: Recent neuroscientific advancements highlight the potential of non-invasive neurorecording methods, such as magnetoencephalography (MEG), electroencephalography (EEG), for the successful detection of auditory attention. This breakthrough has sparked the research enthusiasm of scholars worldwide, particularly in the realm of cognitive controlled hearing aids based on auditory attention detection. Our study introduces a high-accuracy decoding algorithm for auditory spatial attention, leveraging the visibility graph method on multi-channel EEG signals. We propose a network representation method for multi-channel EEG signals based on the similarity features from visibility graph network topologies. The algorithm automatically distinguishes left-ear and right-ear spatial attention during speech auditory activity. When evaluated on the KUL Dataset, our method achieves average attention decoding accuracies of 87.4%-95.7% for durations of 0.5-5 seconds. Comparative results underscore a significant enhancement in auditory spatial attention decoding. Our comparative analysis unequivocally showcases a substantial improvement in auditory spatial attention decoding performance through the adoption of the proposed network representation for multi-channel EEG signals. This innovative approach stands out for its ability to elevate the precision and efficacy of auditory spatial attention decoding.