An algorithm for leukaemia immunophenotype pattern recognition

Abstract

Since leukaemia-specific leucocyte antigen has not been identified to date, the immunological diagnosis of leukaemia is achieved through the application of a wide set of monoclonal antibodies specific for surface markers on leukaemic cells. Thus, the interpretation of leukaemia immunophenotype seems to be a mathematically determined comparison of 'what we found' and 'what we know' about it. The objective of this study was to establish an algorithm for transformation of empirical rules into mathematical values to achieve proper decisions. Recognition of leukaemia phenotype was performed by comparison of phenotyping data with reference data, followed by scoring of such comparisons. Systematic scoring resulted in the formation of new numerical variables allocated to each state, whereas a most significant variable was described as a complex measure of compatibility. A system of recognized states was described by mathematical variables measuring the confidence of information systems, i.e. maximal, total and relative entropy. The entire algorithm was derived by matrix algebra and coded in a high-level program language. The list of the states recognized appeared to be especially helpful in differential diagnosis, occasionally pointing to states that had not been in the scientist's mind at the start of the analysis.

Computer-assisted interpretation of flow cytometry data in hematology

“…Petrovecki et al (16) used methods of pattern recognition for the classification of blood or bone marrow samples into one of 32 states (diseases or normal cell populations). For each of these states, they defined (from literature data) a reference pattern of antigen expression of 49 different antigens where the fraction of antigenpositive cells within one state is classified in a discrete variable by five steps.…”

“…Parts of this work were presented at the 3rd ESACP Conference, Grenoble, France, May [16][17][18][19][20]1994. The authors thank Debra Bickes-Kelleher, for her valuable support in preparing the manuscript.…”

“…Several programs have been introduced for computerassisted analysis and interpretation of both hematological findings (14) and cytometry data (1-3, [5][6][7]9,13,[15][16][17][18][19]. These approaches use statistical methods (5,13,18), artificial intelligence (e.g., neural networks (6,9,13,17), pattern recognition (7,16,18,19), or rule-based systems (1-3, 7,14) to separate several cell populations as a first step toward a diagnosis. Even so, some of these programs (1-3,7, 16,18,19) give the opportunity for the necessary subclassification of hematological diseases from flow cytometry data.…”