Evoked potential processing in neuro-critical monitoring

Evoked potentials (EP) may be monitored to detect events that post high risk to brain in surgical or neuro-critical care environments. We model EP signals by orthonormal basis functions (Fourier, Walsh, etc.). We employ adaptive signal processing algorithms to enhance the signal-to-noise ratio (SNR) and to capture time-varying changes. We analyze time-varying, noisy EP signals by adaptive least mean-square (LMS) and exponentially weighted least squares (EWLS) algorithms to minimize the mean-squared error between each EP sweep and the signal model. A single adaptation parameter controls the convergence and the steady-state SNR in this algorithm. We conducted three sets of experiments to evaluate the performance of these signal processing algorithms, EXPT.1: In anesthetized cats, middle cerebral artery was abruptly occluded and, after some time, released. Transient loss in the somatosensory EP signal and its recovery was captured. Surprisingly, the LMS and the EWLS algorithms exhibit comparable tracking capabilities (LMS, however, requires significantly fewer computations). EXPT.2: Anesthetized cats inspired, in successive stages, N₂O anesthetic, 100% O₂, 8% O₂, and 100% O₂. During the early stages of hypoxia there was a marked transient increase of the energy in the fundamental fraction (fundamental plus the second harmonic), possibly signaling onset of neurological injury. EXPT.3: Human subjects received the anesthetic etomidate. Etomidate elevated the amplitude of the somatosensory EP by 200-500% is less than 100 s. In summary, time-varying changes in the EP signals are captured by adaptive Fourier series modeling (Fourier basis functions and the adaptive LMS algorithm). Anesthetics (etomidate, N₂O) do not alter the shape or the fundamental fraction of the EP signals, but injury to brain does. Transient changes in the fundamental fraction of the somatosensory EP signals, thus, may be used to signal early onset of neurological trauma in critical care settings.