U-Net based classification of flow cytometry data

Abstract

Acute lymphoblastic leukemia (ALL) is the most common type of malignant disease among children. In the case of malignant diseases, abnormal cells can divide uncontrollably and spread to surrounding cells. Chemotherapy is the most common treatment for ALL. Since every patient responds differently to the therapy, it needs to be controlled and individualised. Minimal Residual Disease (MRD) indicates the patient’s response to the treatment. Based on this value the intensity or length of treatment can be modified. Flow cytometry is a laser-based method which can detect MRD. There are several methods for the automatic detection of cancer cell populations based on flow cytometry data with astonishing results but have only one striking disadvantage: the lack of interpretability of the processes. In this thesis, we apply an image segmentation method, the U-Net model, to solve this problem in a visual manner so that all steps of the gating procedure can be easily comprehended. The results meet the performance of the state-of-the-art methods with the additional advantage of easy tractability through the visual representation of the gating procedure.