Algorithm for the Discovery of Clusters in WBC Data

摘要

Analysis of white blood cell flow cytometry light scatter data remains a challenging problem. Conventional methods for analyzing flow cytometry data use two-dimensional scatterplots of multidimensional data. We have developed an automated method to locate and characterize clusters within WBC data. Our method uses the full dimensionality of the data and employs the recursive application of a two-step algorithm Input to the first step is a dataset of cellular events divided into two assumed clusters A clustering algoritfim iteratively refines these initial clusters This algorithm is based upon an extension of the k-means clustering algorithm. If two populations are confirmed, the data for each cluster are passed in turn to the second step a splitting algorithm This algorithm determines the potential for further subdivision of the data When this potential exists, an approximate division of the data is made These new subclusters are passed as input to step one and the process is repeated. The process terminates when either the clustering algorithm converges to a single population or when the splitting algorithm finds no potential for further subdivisions. Using the full dimensionality of the data our method can characterize clusters within WBC data even in cases where these clusters overlap in the standard two-dimenMon.il scatterplots