Using Probability Binning and Bayesian Inference to measure Euclidean Distance of Flow Cytometric data
The British Univesity in Dubai (BUiD)
Flow Cytometry (FCM) is a microscopic technique used in many fields, especially clinical research and health care. Classical analysis of FCM data is done manually in a tedious, error prone process, which is not standardized, not open for re-evaluation and highly dependent on the experience of the analyst. Conventional analysis methods are based on comparisons of univariate or bivariate distributions for one or two channels only, while it is obvious that analyzing flow cytometric data files in a multivariate space would generate more accurate results. For this reason, many studies and researches are directed towards developing a model for automatically analyzing FCM data files, as it is difficult for human analysts to extract clear information from multidimensional data files. The automated analysis of flow cytometric data is challenging due to many reasons especially: the unordered cells across different flow cytometric files and the features are divided across multiple FCS files for the same patient. Many approaches concentrated on resolving either the first or the second challenge, but not both of them. In this thesis, a novel approach is introduced and validated for generating a multivariate flow cytometric data file with N-dimensions, where N is the number of the intended independent measurements. The approach was developed to resolve the main two challenges in flow cytometry – mentioned previously - using concepts of Probability Binning and Bayesian Inference. The approach described in this thesis is validated for classifying normal and leukemia incidence cases. Also, it is validated for classifying different Leukemia types (AML, B-ALL or T-ALL). Experiments show a 100% correspondence between our results and clinical results.
DISSERTATION WITH DISTINCTION
probability binning, Bayesian inference, euclidean distance, flow cytometry