Short Time Fourier and Wavelet Transform for Accelerometric Detection of Myoclonic Seizures

The Short Time Fourier Transform (STFT) and the Continuous Wavelet Transform (CWT) of accelerometer signals measured in patients with epilepsy are analyzed. Characteristics of the spectrogram and scalogram are studied in order to get more insight in how these transforms might be useful to derive suitable features for myoclonic seizure detection. First some artificial elementary patterns that represent some main characteristics of the patterns observed during myoclonic seizures and other movements, are studied. Second the analysis is performed on real patient data. Information from the analysis is used to motivate the use of certain features for detection of myoclonic seizures. Third, for both the STFT and the CWT four feature sets are evaluated in a linear classification setup. Incorporating knowledge from our analysis leads to better detection results than using the spectral power / wavelet coefficients as feature without incorporating this knowledge. This preliminary study shows that both the STFT and the CWT can be valuable for feature extraction for detecting myoclonic seizures. The STFT is more susceptible for false detections.