We compared MRI criteria used to predict conversion of suspected multiple sclerosis to clinically definite multiple sclerosis. Seventy-four patients with clinically isolated neurological symptoms suggestive of multiple sclerosis were studied with MRI. Logistic regression analysis was used to remove redundant information, and a diagnostic model was built after each MRI parameter was dichotomized according to maximum accuracy using receiver operating characteristic analysis. Clinically definite multiple sclerosis developed in 33 patients (prevalence 45%). The optimum cut-off point (number of lesions) was one for most MRI criteria (including gadolinium-enhancement and juxta-cortical lesions), but three for periventricular lesions, and nine for the total number of T2-lesions. Only gadolinium-enhancement and juxta-cortical lesions provided independent information. A final model which, in addition, included infratentorial and periventricular lesions, had an accuracy of 80%, and having more abnormal criteria, predicted conversion to clinically definite multiple sclerosis strongly. The model performed better than the criteria of Paty et al. (Neurology 1988; 38: 180-5) and of Fazekas et al. (Neurology 1988; 38: 1822-5). We concluded that a four-parameter dichotomized MRI model including gadolinium-enhancement, juxtacortical, infratentorial and periventricular lesions best predicts conversion to clinically definite multiple sclerosis.
Frederik Barkhof (1997) studied this question.