Mixtures of normal distributions in the problem of reference points search using myograms

The paper is devoted to investigation of myogram probability characteristics. This signal represents a record of electrical activity produced by skeletal muscles. It is widely used in medical research, including the determination of reference points in the problem of localization of functional brain areas. The authors propose to take finite scale-location mixtures of normal distributions as a mathematical model of myogram noise. Separation of mixtures is solved by the stochastic EM (expectation–maximization) algorithm and obtained data are used to reveal start points for movements using CUSUM statistics. Finally, the authors compare the new algorithm with the method based on window variance thresholding, which is already used in the MEG center.