The existing process of leukemia diagnosis can be overly subjective, inefficient, and costly. Measurements of individual cells, done using flow cytometry, are typically interpreted using manual gating processes; these processes introduce human bias. This paper proposes a reliable, cost-effective, and quick diagnostic process using machine learning (ML). These ML-powered systems automate the gating of flow cytometry data to improve the accuracy of diagnosis. Through leveraging supervised and unsupervised ML models, such as logistic regression, random forests, and convolutional neural networks (CNNs), this study aims to produce a neural network capable of diagnosing leukemia from FCM data accurately. To improve the efficiency of the system, UMAPs are utilized for dimension reduction and data visualization. 

This project follows a pipeline involving data preprocessing, model creation, training, and post-processing steps to ensure accurate and interpretable results. Strengths and weaknesses, such as data acquisition, training time, and result interpretation, are also considered; alternative designs are analyzed for feasibility and effectiveness. We present a comprehensive framework for the development of a neural network-based tool to improve the accuracy and efficiency of leukemia diagnosis.

Acknowledgments:

Special Thanks to our mentors Dr. Yu “Max” Qian, Dr. Alyssa Taylor Amos, and Noah Mehringer. This project would not have been possible without their continued guidance. 

Credit: Samuel Wang