AUTOMATED CLASSIFICATION OF FLOW CYTOMETRIC THREE-COLOUR IMMUNEPHENOTYPES FROM BLOOD LEUKOCYTES AND BONE MARROW OF LOW- AND HIGH-GRADE NON-HODGKIN LYMPHOMA AND CHRONIC LYMPHATIC LEUKEMIA PATIENTS.

G. Valet 1), G. Schmidtke 2), U. Schmücker 2), G. Brittinger 2), H.G. Höffkes 2),

1) Arb. Gruppe Zellbiochemie, Max-Planck-Institut für Biochemie, D-82152 Martinsried,

2) Div. Hematology, Dep. Medicine, University of Essen, D-45122, Essen, Germany.

Three colour immunephenotypes provide information on lympho-, mono- and granulocyte compartments by the forward/sideward scatter (FSC/SSC) histogram as well as by quadrant evaluation of the FITC/PE, PE/PerCP and FITC/PerCP histograms. Typically, the percent lymphocyte frequencies in 3x4=12 quadrants or in case of absolute counts 2x12=24 numbers are extracted.

This information is not exhaustive because neither fluorescence intensities, fluorescence ratios, relative fluorescence densities on the cell membrane, nor mono- and granulocyte informations are evaluated. The CLASSIF1 software (Ann.NY Acad.Sci. 677,233-251 (1993), Partec, Münster, Germany) determines all these parameters in addition to the above percent and absolute cell count values. This leads to the introduction of 222 values into a database for one three colour measurement or of 1110 database columns for five measurements.

CD45/14/20, CD8/4/3, kappa/CD19/5, lambda/CD19/5, CD10/23/ 19 list mode files (BD-FACScan) of blood leukocyte and bone marrow samples of chronic lymphatic leukemia (CLL, n=14), hairy cell leukemia (n=8), centrocytic- centroblastic lymphoma (n=2), centroblastic lymphoma (n=2), immunocytoma (LP-1C, n=9), B-cell lymphoma (n=1) as well as of normal persons (n=15/15) were available for this pilot study in addition to blood leukocyte samples from unspecific leukocytosis (n=12), centrocytic lymphoma (n=1) and low grade lymphoma (n=1).

The CLASSIF1 program iteratively eliminates database columns which do not improve the identification result in the confusion matrix. The confusion matrix is constituted by the known clinical diagnoses and the initially unknown flow cytometric classification. Increase of the diagonal values of the confusion matrix towards the 100% correct identification of clinical samples by the standardized, laboratory and instrument independent classifier is the goal of the iteration process.

Individual patient samples were correctly classified in between 84% to 100% of the cases. The clinically difficult distinction between CLL and LP-IC was possible for 100% of the patients. In addition 9 of 10 clinically unassignable patients were clearly assigned to CLL or LP-IC. The information for these decisions derived mostly from a combination of 25-35 lympho-, mono- and granulocyte (blast) parameters.