Study On The Latency Change Of Evoked Potential In α-stable Process Model

Evoked potentials (EPs) are bioelectric signals generated by central nervous system when it is stimulated by well-defined external events. The EP latency and the latency changes (delays) indicate the actual conduction and delay of the neurological system. So the latency change detection of EPs is of special interest in many clinical applications, such as diagnosis of the injury and pathological changes in the nervous system.Traditional EP analysis methods often assume that electroencephalogram (EEG) and other noises are Gaussian distributed. However, result studies showed that the α-stable process model may be more appropriate than the traditional Gaussian model for describing the noise encountered in EP analysis for certain experimental conditions. Since an a -stable process has infinite second order moments, the performance of some traditional latency detection algorithms on the assumption that the second order moments are finite, such as LMS and DLMS degenerate seriously. So developing robust EP analysis algorithms under a -stable noise condition is of special interest in the diagnosis of the injury and pathological changes in the nervous system.For the latency change detection of evoked potentials, the effect of a -stable noise can be restrained effectively through implementing appropriate non-linear transform to the error function. This paper proposes a new method for EP latency change detection based on the principle of μ -law compression. The result demonstrates that the new method can adjust the parameters dynamically according to the noise variety. The improved performance and robustness over exiting algorithms are also demonstrated under the different noise conditions through computer simulation.In clinical application, we need to apply a sweep of the normal EPs as a reference signal of adaptive time delay estimation. But the normal EPs without latency changes is difficult to obtain from one who is suffering the disorder or injury in the CNS. This thesis proposes a method of EP signal construction based on the character of the spectrum function. The result demonstrates that the constructed EP signal is effective as a reference signal for the latency change detection of evoked potentials.This thesis also employs a BP neural networks (NN) to detect known signals under the a -stable noise condition. Training this NN detector at some specified probability of false alarm and then adjusting the weights of the bias nodes, we can acquire the test statistics. The result demonstrates that under the α-stable noise condition, the NN detector detects the signals effectively.