Objective evaluation of two markers of HIV-1 infection (p24 antigen concentration and CD4+ cell counts) by a self organizing neural network.

The aim of the present work is to obtain groups of patients with similar profiles of p24 antigen concentration and of CD4+ cell counts. These two markers were chosen because their evaluation represents a significant step in the clinical follow up of HIV-1 infected subjects. The detection methods for p24 antigen concentration and for CD4+ cell counts are well assessed and guarantee easy reproducibility of data obtained in different laboratories. A set of observations with the same time intervals were derived from a continuous function obtained for each patient by a back-propagation neural net trained on the raw data from the patient. The classifications were obtained by a Kohonen neural net trained in three ways: with p24 antigen profiles only, with CD4+ cell count profiles only and with both sets of profiles. The results show that the clustering fashion of the two parameters closely resembles the clustering fashion of CD4+ only, rather than that of p24Ag, both with reference to cluster formation and with reference to distances between clusters.