Toward leveraging big value from data: chronic lymphocytic leukemia cell classification

Abstract

The goal of Big Data analysis is delineating hidden patterns from data and leverage them into strategies and plans to support informed decision making in a diversity of situations. Big Data are characterized by large volume, high velocity, wide variety, and high value, which may represent difficulties in storage and processing. Research on Big Data repositories has contributed promising results that primarily address how to efficiently mine a variety of large volume of structured and unstructured data. However, innovative insights can emerge while leveraging the value characteristic of Big Data. In other words, any given data can be big if analytics can draw a big value from it. In this paper, we demonstrate the potential of five machine learning algorithms to leverage the value of medium size microscopic blood smear images to classify patients with chronic lymphocytic leukemia (CLL). The maximum majority voting method is used to fuse the predications made by the five classifier models. To validate this work, 11 CLL patients are refereed by flow cytometry equipment and the results are compared to the proposed classifier model. The proposed method proceeds through a sequence of steps while working with the lymphocyte images: it segments the lymphocyte images, extracts/selects features, classifies the selected features using five classifiers, and calculates the majority class for the test image. The proposed composite classifier model has an accuracy of 87.0%, true-positive rate of 84.95%, and 10.96% false-positive rate and can correctly identify 9 out of 11 patients as positive for CLL.