THE WAVELET ANALYSISOF EVOKED POTENTIALS

Abstract: In this paper,we use the orthogonal wavelet based on 3-order splines function to processthe somatosensory evoked potentials and cognitive event-related potentials. The experiments onhuman subjects have showed that this method is valuable and practicable.INTRODUCTIONEvoked potentials(EP' s) are electrical responses of the central nervous system to stimulusapplied in a controlled manner. They can not only be used to diagnose the diseases on the sensorypathway and brain but also be frequently monitored intraoperatively to assess the effects ofsurgery or to detect unexpected neurological injury [1][2]. But the Signal-to-Noise ratios (SNRIs) of EP' s are vary low. The: are many methods in order to improve the SNR' S and identify theEP' s signal perfectly with less stimuli such as ensembl averaging,post-wiener filtering,adaptedfiltering and parametric modeling and even single stimulus. The purpose of this paper is to applyrnultiresolution wavelet analysis method for identification of short-latency somatosensory evokedpotentials and long-latency evoked--related potentials.METHODMultiresolution Wavelet analysis and pyramid algorithm was constructed by Mallat [3] andwidely used in signal processing pattern recognition and image processing. Here,we applyed thisto decompose EP' S signal into different frequency bands and reconstructed them with time-varying weight method.Based on 3-order splines functions,we constructed orthogonal wavelet and scale function. According to the dual-scale function,we could get transform function h and its mirror filter g,theyare related by gn(- l)nhn+1. More detail could see Mallat' theory[3]. Through fast pyramid algorithm,the EP' S signal can be decomposed into detail and coarse parts on a dyadic scales by scalefunction like this Here Cj is coarse signal and Dj is detail signal. C, is signal on scale 0 and is the same as originalsignal. With j increasing,the EP' S signal were projected to different frequency bands. The reconstructed function for signal on j scale to j-1 scale isAccording to the character of EP' S signals and their decomposition on different scales ,we choosethe following signal processing principle.To somatosensory EPI S (SEP),we choose coarse and detail signal on scale 6 to reconstructsignalS(t) ~W, (t) o C,(t) +WZ(t) o D,(t)W, and W, are time--varying weight and are decided by the amplitude of detail and coarse signalson scale 6.To Event--related optentials (P300 waveform),because the noise is mainly on high frequencyband gwe choose coarse signal of scale 6 to reconstruct signal.RESULTThere are ten subjects taking part in the experiment. The instrument is SPECTRUM32 ofCADWELL corporation. To SEP signals,The medial nerve of left wrist was stimulated,activeelectrode was attached to Cll (3 cm lateral to the midline and posterior to the coronal suture).reference electrode at Al (left ear) and ground electrode at Fpz according to 'the international 10-20system of electrode placement.Figurel is the original SEPI S signal and processed signal, (a) is the SEPI S signal averaged by250 times. (b) is the signal averaged by 30 times and (c) is the signal of (b) be processed. Fromthe figure,we could see that the method is effectively,noise ate inhibited greatly and the waveform of (c) is almost the same as the signal (a).Figure 2 .The EPRI S signal and processed signalTo ERP,active electrode was attached to Cz.reference electrode at Al and ground electrodeat Fpz. The stimulus is auditory and the probability of target stimulus is 20 percent.FigureZ is the EPR signal, (a) is the signal average 40 times, (b) is the single trial signal and(c) is the processed signal of (b). From the figure, we could say that multiresolution waveletanalysis can get single trial signal of ERP. The signal (c) is almost the same as signal of (a).From the experiments,we could say that the time--varying weighted multiresolution waveletanalysis method is suitable and practicable and may be vary useful in identification of single trialEPI s signal.