An automated data analysis strategy to identify and classify hematologic neoplasms (HEM3P.287)

Abstract

Abstract Cell immunophenotyping using flow cytometry has historically been a powerful tool in screening and diagnosing hematologic neoplasms. The capability of flow cytometers to measure multiple markers simultaneously allows detecting and quantifying normal and aberrant immune cell populations in blood samples. However, analysis of multidimensional flow cytometry data remains cumbersome and limited to serial analysis of bivariate plots. The slow and visually-dependent method relies on an expert’s expertise to identify and gate immune cell populations based on the expression and intensity of targeted protein marker staining. This conventional gating method is subjective, time consuming, and hardly takes advantage of all the dimensionalities of the data. An automated data analysis strategy to identify abnormal populations and classify samples in a gating-free manner has been developed. This automated analysis contains algorithms to extract features from high dimensional data, merge features and create profile for each sample, and classify samples based on established profiles. This generalized analysis strategy could be applied to analyze multidimensional data. An application on hematologic neoplasm classification is presented here. More than 100 samples, including health and patient samples, stained with a same eight color antibody cocktail were acquired using a BD FACSCanto II, and was used to show the accuracy of the automated analysis method is higher than 90%.