The notion of a "myopathic" or "neuropathic" electromyogram (EMG) is usuall
y based on qualitative visual and acoustical impressions. Conventional quan
tification defines abnormality but not diagnosis, which requires interpreta
tion of patterns of change, Discriminant analysis is a model for this multi
variate decision. It tells how probable it is that a motor unit potential (
MUP) comes from a normal, myopathic, or neuropathic muscle. Accumulation of
single MUP information by a sequential Bayesian algorithm produced diagnos
tic probabilities above 0.95 in 91% of all muscles (223 biceps brachii musc
les from 80 patients with motoneuron disorders, 56 patients with neuropathi
es, 71 patients with myopathies, and 34 controls). Two muscles from patient
s with neurogenic disorders were misclassified as "myopathic." Misclassific
ation was more frequent only in myositis (4 of 28 muscles) and in oculophar
yngeal muscular dystrophy (2 of 4 muscles). MUP discriminant classification
was as sensitive as, and more specific than, conventional quantitative EMG
, which discriminated between myopathic and neuropathic in only 22% of the
muscles. This rate was 59% for discriminant analysis. As a knowledge-based
expert system, MUP discriminant analysis successfully distinguishes between
myopathic, neuropathic, and unclassifiable MUP samples. It discloses more
information than conventional quantitative MUP analysis. (C) 1999 John Wile
y & Sons, Inc.