Time-Frequency Domain Classifier for Propofol-Mediated Unconsciousness.

Wavelet analysis is a powerful tool in signal processing, enabling precise time-frequency localization that is well-suited for non-stationary signals such as electroencephalograms (EEG). We developed a wavelet-based classifier to assess the level of propofol-mediated unconsciousness. For feature extraction, we used a 6-level Discrete Wavelet Transform (DWT) decomposition followed by simple time domain transformations. The multi-class gradient-boosting classifier generated probability estimates for three classes: conscious, unconscious, and burst suppression. For each window, a continuous class estimate (CCE) was calculated based on these probabilities, and the class with the highest probability was selected as the predicted label. We compared the performance of the DWT-based classifier with other methods that use frequency and time-frequency feature extraction in terms of accuracy and computational efficiency. The DWT-based classifier achieved 92.9% accuracy, comparable to other methods, while providing improved detection of burst suppression events. Additionally, an optimized version of the DWT-based classifier, using a reduced set of features, achieved a similar accuracy of 92.3%, while offering improvements in computational efficiency.