Enhanced time-frequency features for neonatal EEG seizure detection

Time-frequency based methods have been proved to be superior to other methods in analysing neonatal EEG. This is due to the fact that newborn EEG is nonstationary and multicomponent. This paper presents an approach for improving the performance of the EEG seizure detection technique previously introduced by the authors. The proposed approach utilizes the SVD-based technique for both enhancing the time-frequency representation of the signal and extracting EEG seizure features. Enhancing the time-frequency representation leads to improvement in the quality of the extracted feature. To extract the features the estimated distribution functions of the singular vectors associated with the time-frequency representation of the EEG epoch are used to identify the patterns embedded in the signal. The estimated distributed functions related to the seizure and nonseizure epochs were used to train a neural network to discriminate between seizure and nonseizure patterns.