Computer-assisted diagnosis of hematological malignancies using a pattern representation of flow cytometry data

Describes an effort to rapidly classify multidimensional probabilistic data. The effort models quantized process data provided by smart sensors with biological immunophenotypic data as measured with a flow cytometer. The immediate goal is the classification of hematologic malignancies such as leukemias and lymphomas and the monitoring and evaluation of chronic illnesses such as AIDS and cancers. The diagnosis and classification of hematological malignancies is a multivariate decision involving a number of individual parameters. The authors present an alternative approach to clinical flow cytometry by addition to the instrument of the tube identification parameter unit which was designed and constructed both to reduce sample identification error and to combine the entire phenotypic assay into a single data file without loss of information. The resulting file structure allows data analysis of uniquely created data arrays the authors have termed ´phenograms´. This resulting pattern is analogous to a collection of values defining the degree of membership to multiple fuzzy sets. The phenogram pattern, along with other relevant patient data, can then be provided as input for a neural classification system for diagnostic assistance.