Detection and classification in uterine electromyography by multiscale representation

The purpose of this study was to detect the uterine contraction and other events in uterine electromyography (EMG). Classical statistical algorithms of change detection use a criterion based on signal variance in a sequential manner. The idea of the present work is to generalize these algorithms in a multidimensional way using a multiscale signal decomposition. In this case, the classical algorithms are modified by use of a criterion based on covariance matrix estimates. The isolation is achieved by a multi-hypothesis test. Thus, the successive events are detected by a single multidimensional detector and classified regarding their covariance matrix. This approach leads to different detection and isolation delays, the former being defined by the detection threshold, the latter depending on the estimation time of the covariance matrix