The predictive power of MRI criteria in the diagnosis of immune reconstitution inflammatory syndrome by the comparison of classification using the decision tree method and ROC analysis

. Immune reconstitution inflammatory syndrome involving the central nervous system (CNS-IRIS) is a dangerous complication in HIV-infected patients on antiretroviral therapy (ART). The radiologic features of this syndrome have been little studied and are presented in isolated works. The diagnosis is difficult because there are no generally accepted criteria for IRIS. Our study is devoted to radiology of IRIS. Based on the results of brain MRI, together with clinical and laboratory data, MRI criteria for IRIS were formulated. Purpose and goals . To determine the prognostic value of MRI signs of CNS-IRIS using in a cohort of HIV-positive patients with neurological symptoms. Materials and methods . The analysis includes data from 68 HIV-infected patients who underwent brain MRI. In 14 of them were diagnosed IRIS with involvement of the central nervous system. To determine the diagnostic efficiency of the formulated MRI criteria, the STATISTICA program was used, decision trees were built, and a ROC analysis was performed. Results . Five decision tree models were built with different predictive values. The models took into account the categorical predictors (MRI criteria) in different order and quantity. The best performance has model #5, which can be considered a clinically useful predictive model. Conclusion. Brain MRI is an essential diagnostic step in HIV-infected patients on ART. It is necessary to expand the indications and conditions for radiological studies of the brain in patients with suspected immune reconstitution inflammatory syndrome.