Bayesian statistics as applied to multiple sclerosis diagnosis by evoked potentials

This study deals with the multiple sclerosis diagnosis by visual, somatosensory and brain stem auditory evoked potentials. By a statistical investigation of experimental data, four items have been selected as being both discriminative and independent: P100, N20, P40 latencies and the auditory central conduction time. The Bayesian approach was then attempted, i.e. the marginal density distributions of every item, and the joint density distribution functions have been established within both normal and multiple sclerosis population. No loss function consideration have been taken into account, as the final diagnosis is established by a neurologist who disposes of additional clinical and paraclinical information. The discrimination, as given by posterior Bayesian probability, is very satisfactory since the ratio of recognized multiple sclerosis among MS patients is over 90%, while no normal subject is diagnosed as MS.