To investigate whether peripheral immune abnormalities are associated with brain inflammation in multiple sclerosis, and whether differences in MRI activity are paralleled by changes in leukocyte composition, we conducted a prospective longitudinal study in patients at their clinical onset. Twenty patients presenting a first inflammatory event in the central nervous system suggestive of multiple sclerosis underwent, every 45 days for one year, immunophenotyping of 98 blood cell subsets together with brain MRI and clinical evaluation. Six patients showed intense MRI activity, six patients did not display MRI activity, while the remaining 8 patients had low (i.e. intermediate) MRI activity during the follow-up. Our results show that MRI-active and MRI-inactive patients display significant differences in ten lymphocyte subsets. Among these, there are both effector (CCR7-CD45RA-CD4+ alphabeta T cells, CCR5+ gammadelta T cells) and regulatory (DN CD28+ alphabeta T cells and CD25+CD8+ alphabeta T cells) lymphocytes pertaining to the innate and the acquired arms of the immune system. Moreover, these differences were, upon employment of a class prediction procedure based on "support vector machines" algorithm utilizing leave-one-out cross validation procedures, able to correctly assign patients to their respective MRI activity group. All 6 MRI-active and 6 MRI-inactive patients were correctly classified, and, upon application of a class prediction model in an unsupervised manner to the 8 patients with intermediate MRI activity, 6 were predicted as MRI-active and 2 as MRI-inactive patients. Also, when the mean values of the first three time points (T0, T1 and T2) were used for the prediction of all patients, the selected lymphocyte subsets correctly classified 90% of patients. Sensitivity was 91.7% and specificity was 87.5%. These results provide evidence showing that brain inflammation in multiple sclerosis is associated with distinct changes in peripheral lymphocyte subsets, and raise the possibility that the identified subsets may, after adequate validation, assist in the prediction of MRI activity in the early stages of multiple sclerosis.