To test the reliability of four previously proposed MRI criteria for t
he diagnosis of MS, we reviewed 1,500 consecutive brain scans for the
presence, number, size, and location of areas of increased signal (AIS
) on proton-density and T2-weighted images, unaware of the patients' c
linical presentations and ages. This series included 134 subjects with
a clinical diagnosis of MS. Relying exclusively on the presence of at
least three or four AIS for a positive diagnosis of MS resulted in hi
gh sensitivity (90% for three AIS and 87% for four) but inadequate spe
cificity (71% for three AIS and 74% for four) and positive predictive
value (23% for three AIS and 25% for four). If one of these lesions wa
s required to border the lateral ventricles, specificity was 92% and p
ositive predictive value was 50% at a sensitivity of 87%. Using the Fa
zekas criteria (at least three AIS and two of the following features:
abutting body of lateral ventricles, infratentorial lesion location, a
nd size >5 mm) led to a further highly significant improvement of spec
ificity (96%; p = 0.0000) and increase of the positive predictive valu
e (65%) at the expense of a less significant decrease in sensitivity (
81%; p < 0.01).