On the application of model based distance metrics of signals for detection of brain injury

In the basic and clinical research on brain's response to injury, electrical signals from the brain, namely EEG, is useful in providing an immediate signaling of the dysfunction. However, EEG signals have proven to be difficult to analyze and interpret due it its complex signal characteristic. There is a critical need for developing robust and reliable measures that can be correlated with injury as well as survival. In this paper, we address a unique problem of characterizing quantitatively the electrical measures of brain injury for analysis of brain activity in animal and human subjects. The key objective is to model EEG spectra and its features so that signaling changes due to injury can be discovered. We do so with the method of autoregressive modeling and dominant frequency analysis. The trends in the electrical signaling following injury and following resuscitation are modeled using the cepstral distance derived from the AR model.