Analysis of biomedical signals by means of linear modeling.
Abstract
The recording and subsequent analysis of electrical signals of physiological origin constitutes an important aspect of current biomedical research. A versatile method for the analysis of such signals is based on linear, i.e., autoregressive (moving average) modeling. These techniques are based on fitting a hypothetical model to the signal under observation. These models are capable of generating the original signal by a linear combination of past observations and past and present noise samples. High resolution spectral estimates can be obtained in this way. Also, the often small number of model coefficients offer a concise description of the signal and may be used for classification purposes. Other applications entail the detection of nonstationarities, data-compression, and signal enhancement. In this review, linear modeling methods for the analysis of electroencephalograms, electro- and phono-cardiograms, electromyograms, and gastrointestinal signals are surveyed.