Evaluation of Filtering Techniques applied to simulated Electroencephalogram signals for Visual Evoked Potential Detection

Résumé 0

Introduction: In this paper, it is presented an assessment of different windowing techniques and denoising methods applied to detect the Visual Evoked Potential (VEP) contained into the electroencephalographic (EEG) signals. The objective is to analyze advantages and disadvantages of each technique performance to identify correctly the VEP waveform and to quantify the latency which is a parameter commonly used as an indicator of ocular diseases. Method: Assessment of techniques was performed considering rigorous controlled conditions on the signal set, to assure that obtained results were linked just to the technique performances and avoid any undesirable effect due to the natural interference of acquisition of signals. For this reason, a simulated signals set was created based on a typical VEP waveform commonly registered around 100ms after an external stimulus was applied in the routine clinical study. Additionally, two evaluation stages were considered: fixed latency parameters stage and random latency parameters stage. Results: The best results without filter was with rectangular window and the use of elliptic filter can help to extract the VEP with a rectangular window too. For wavelet denoising the best result is Biorthogonal 2.6 wavelet with a Hamming window. Discussion or Conclusion: Five parameters were proposed to assessment the VEP extraction performed: signal noise ratio (SNR) mean square error (MSE), average latency (AL), latency standard deviation (LSD) and latency correlation variance (LCV) as representative factors to be considered on evaluations of VEP extraction methods. SNR and MSE were focused to assess the level of noise that remain in signal after windowing & denoising method was applied. In the other hand AL, LSD and LCV were oriented to evaluate the impact of the method on the VEP latency estimation.