Multivariate discriminant analysis of the electromyographic interference pattern: statistical approach to discrimination among controls, myopathies and neuropathies.

Abstract

The stepwise linear discriminant analysis method is used to develop optimal combinations of features measured from the electromyographic interference pattern, with the aim of minimising the misclassification rate in controls while maximising the correct classification rates in patients with disease. This discriminant analysis among multiple groups leads to the determination of the optimal discriminating surface in a multivariable space and can also produce a severity of disease likelihood index. Applying these combinations of features to 186 studies performed in the biceps muscle, 81% of all studies are accurately classified as being normal, myopathic or neuropathic. An algorithm to perform this stepwise multigroup linear discriminant analysis is described.