ON DISCRIMINATION OF (STATIONARY) STOCHASTIC-PROCESSES

An essential basis of medical diagnosis are biopotentials obtained from the body-surface of the patients. If these time-functions are to serve for computer-aided diagnostics (using discrimination procedures) the known methods fail because of the existing small sample sizes for a large number of features (amplitudes). In the paper a method is given allowing discrimination of biopotentials also with such an extreme ratio of the number of features to the sample size. Furthermore, univariate tests are provided, enabling a decision to be made on whether the possibly arising distinctions may be attached to the mean values and/or the spectra of the underlying potentials. Clinical applications from the field of otological diagnostics demonstrate the usefulness of the described methods and show in particular their superiority compared with the known linear discriminance analysis.