ERPs-Based Attention Analysis Using Continuous Wavelet Transform: The Bottom-up and Top-down Paradigms

摘要

Evoked Related Potentials (ERPs) analysis for distinguishing between bottom-up and top-down attention, using 256-channel EEG signals obtained by measurements on humans, is investigated here. The three main ERPs, i.e., N170, P300, N400 were obtained after appropriate time windows selection for different response peaks and averaging EEG waveforms from all trials for each channel. Following that, Continuous Wavelet Transform (CWT) was applied on each ERP waveform and a vector of six morphological CWT-based features was constructed and fed to two well-known classifiers, i.e., SVM and k-NN. The selected ERPs were drawn from those channels where they are known to be observed more frequently. The experimental results have shown that P300 provides higher classification rates than N170 and N400, reaching a classification accuracy of 76%. Moreover, SVM and k-NN showed similar performance, with the latter being slightly more efficient. Finally, gender factorization of data contributed to a maximum classification accuracy of 80%. The proposed analysis paves the way for better understanding of the activity of the brain in different attention scenarios as reflected in the CWT domain, exploring the time-frequency characteristics of the related ERPs, contributing to the detection of potential attention disorders.