Chronic lymphocytic leukemia patient classification methodology through flow cytometry analysis

Abstract

Flow cytometry (FC) is widely used for diagnostic purposes in clinical practice. This analysis typically aims at clustering cellular events according to their biological characteristics, known as gating, and then use the selected clusters in order to conclude about clinical outcomes. As each step of this process is highly subjective, various proposed methods have attempted to automate each step of the procedure separately, but any method has been proposed in order to automate the whole diagnostic process. We constructed a tool that simulates the experts decisions during the whole process in order to conclude if a sample is pathologic or not (`healthy'). We used flow cytometric data from 10 individuals with a diagnosis of chronic lymphocytic leukemia (CLL) from a panel that produces 7 files for each sample. With the help of the present tool we were able to identify whether the analysis of the tested sample confirms the diagnosis of CLL, thus successfully reproducing the experts' decisions at each step of the diagnostic workflow. The validation was conducted by experts against the traditional manual procedure. The proposed methodology is the first attempt to automate the entire process, which is a prerequisite for a fully automated diagnostic system that would ensure objectivity to the clinical diagnostic procedure. The experimental results presented herein show that our proposed new technique has satisfying performance at each level of evaluation.