Automated classification of lymphoid cells.
J Juan, F Sigaux, Georges Flandrin · PubMed · 1985
A morphometric study using a commercially manufactured automated analyzer was carried out on 18,742 circulating blood lymphoid cells obtained from patients suffering from a lymphoid blood disorder and from healthy controls. A sequence of statistical procedures was applied to the multiparameter morphometric data, enabling us to divide up the cell population into classes and to represent a given cell sample by a vector whose coordinates correspond to the distribution of the cells among the different classes (the lymphoid differential count). Automatic classification of the samples was carried out using this model. The groups obtained roughly matched the cytologic diagnosis, even though no diagnostic indications were added to the morphometric data. The lymphoid differential also permits good discrimination between chronic lymphoid leukemia and acute lymphoblastic leukemia. The method, suitable for large data sets, performs a reduction of the data that, because of its flexibility, achieves a compromise between loss of information and ease of application. It appears well suited for multiparameter studies of cell subpopulations whose morphologic features form a continuum and for which prior delimitation of classes is difficult, poorly reproducible or artificial.