Leukemia is the most frequent malignancy in children and adolescents, with acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML) as the most common subtypes. Minimal residual disease (MRD) measured by flow cytometry (FCM) has proven to be a strong prognostic factor in ALL as well as in AML. Machine learning techniques have been emerging in the field of automated MRD quantification with the objective of superseding subjective and time-consuming manual analysis of FCM-MRD data. In contrast to ALL, where supervised multi-class classification methods have been successfully deployed for MRD detection, AML poses new challenges: AML is rarer (with fewer available training data) than ALL and much more heterogeneous in its immunophenotypic appearance, where one-class classification (anomaly detection) methods seem more suitable. In this work, a new semi-supervised approach based on the UMAP algorithm for MRD detection utilizing only labels of blast free FCM samples is presented. The method is tested on a newly gathered set of AML FCM samples and results are compared to state-of-the-art methods. We reach a median $F_1$-score of 0.794, while providing a transparent classification pipeline with explainable results that facilitates inter-disciplinary work between medical and technical experts. This work shows that despite several issues yet to overcome, the merits of automated MRD quantification can be fully exploited also in AML.

While model architectures and training strategies have become more generic and flexible with respect to different data modalities over the past years, a persistent limitation lies in the assumption of fixed quantities and arrangements of input features. This limitation becomes particularly relevant in scenarios where the attributes captured during data acquisition vary across different samples. In this work, we aim at effectively leveraging data with varying features, without the need to constrain the input space to the intersection of potential feature sets or to expand it to their union. We propose a novel architecture that can directly process data without the necessity of aligned feature modalities by learning a general embedding space that captures the relationship between features across data samples with varying sets of features. This is achieved via a set-transformer architecture augmented by feature-encoder layers, thereby enabling the learning of a shared latent feature space from data originating from heterogeneous feature spaces. The advantages of the model are demonstrated for automatic cancer cell detection in acute myeloid leukemia in flow cytometry data, where the features measured during acquisition often vary between samples. Our proposed architecture’s capacity to operate seamlessly across incongruent feature spaces is particularly relevant in this context, where data scarcity arises from the low prevalence of the disease. The code is available for research purposes at https://github.com/lisaweijler/FATE.