Parametric modeling of somatosensory evoked potentials using discrete cosine transform.

This paper introduces a parametric method for identifying the somatosensory evoked potentials (SEPs). The identification was carried out by using pole-zero modeling of the SEPs in the discrete cosine transform (DCT) domain. It was found that the DCT coefficients of a monophasic signal can be sufficiently approximated by a second-order transfer function with a conjugate pole pair. The averaged SEP signal was modeled by the sum of several second-order transfer functions with appropriate zeros and poles estimated using the least square method in the DCT domain. Results of the estimation demonstrated that the model output was in an excellent agreement with the raw SEPs both qualitatively and quantitatively. Comparing with the common autoregressive model with exogenous input modeling in the time domain, the DCT domain modeling achieves a high goodness of fitting with a very low model order. Applications of the proposed method are possible in clinical practice for feature extraction, noise cancellation and individual component decomposition of the SEPs as well as other evoked potentials.