oa Newborn EEG Seizure Detection Using Time-Frequency Matched Filtering

Background: The analysis of Electroencephalography (EEG) signals acquired from epileptic babies shows that seizures can be modeled as piecewise linear frequency modulated (LFM) signals. This fact motivated the use of time-frequency matched filters (TFMFs) for seizure detection in newborn EEG. A TFMF is characterized by a unique test statistic, which is found based on the time-frequency (TF) correlation between the signal under analysis and a template. The test statistic is compared with a threshold to determine the presence or absence of the template in the signal under analysis.

Objectives: We present two seizure detection algorithms based on the general class of TFMFs and an improved algorithm in the ambiguity domain and evaluate their performance using real EEG signals.

Methods: The method includes the following stages:

Based on TF analysis of newborn EEG, we create a template set containing M piecewise LFM signals with L pieces and slopes.

We define test statistics based on the TF correlation between the EEG signals under analysis; we use the Wigner-Ville distribution (WVD) and other quadratic Time-Frequency Distributions.

The test statistics are compared with a predefined threshold.

Results: We evaluated the performance of the proposed method using a database of newborn EEG signals. For each method, we found the area under the receiver operator characteristic curve (AUC) as the performance criteria. All the methods detected seizure accurately with AUC more than 0.9.

Conclusions: This work shows that TFMFs can detect seizures in newborn EEG with a very high accuracy. The optimization of the parameters of the TFDs and the use of fast and memory efficient algorithms for computing TFDs can improve the performance of the methods.