Classification and regression trees for bone marrow immunophenotyping

Methods are needed to assist with automating three-color flow cytometric immunophenotyping of bone marrow from leukemia patients. Described is a method in which a normal bone marrow data set is used as a template against which to compare leukemic bone marrow data sets. This template is obtained using techniques of cluster analysis and cluster editing. Leukemic cells often inappropriately express antigens and appear in a different part of the multivariate data space than normal cells. To recognize the cells exhibiting inappropriate antigen expression, an artificial cluster of "cells" is added to the normal template. The "cells" in this cluster fill the space not occupied by normal cells. Classification and regression tree (CART) analysis is used to train a classifier that can then be used to isolate the major cell types and the inappropriate expression cells in a leukemic bone marrow specimen.