Hematologist‐Level Classification of Mature B‐Cell Neoplasm Using Deep Learning on Multiparameter Flow Cytometry Data

Abstract

The wealth of information captured by multiparameter flow cytometry (MFC) can be analyzed by recent methods of computer vision when represented as a single image file. We therefore transformed MFC raw data into a multicolor 2D image by a selforganizing map and classified this representation using a convolutional neural network. By this means, we built an artificial intelligence that is not only able to distinguish diseased from healthy samples, but it can also differentiate seven subtypes of mature B-cell neoplasm. We trained our model with 18,274 cases including chronic lymphocytic leukemia and its precursor monoclonal B-cell lymphocytosis, marginal zone lymphoma, mantle cell lymphoma, prolymphocytic leukemia, follicular lymphoma, hairy cell leukemia, lymphoplasmacytic lymphoma and achieved a weighted F1 score of 0.94 on a separate test set of 2,348 cases. Furthermore, we estimated the trustworthiness of a classification and could classify 70% of all cases with a confidence of 0.95 and higher. Our performance analyses indicate that particularly for rare subtypes further improvement can be expected when even more samples are available for training.

Expert-independent classification of mature B-cell neoplasms using standardized flow cytometry: a multicentric study

“…By definition, the algorithm excludes that more than 1 diagnosis is made in each case. In contrast to other algorithms, 17,21 this rigid approach resulted in a sizeable number of cases that are not assigned to any particular lymphoma entity, thus reducing sensitivity. However, we feel that the inclusion into the “not classified” category is of great clinical utility as it informs of the necessity for ancillary testing by histology, cytogenetics, and/or molecular biology approaches.…”

“…This accuracy equals previous single-center observations for CLL and HCL but exceeds published data on FL and MCL. 17,21 High PPVs in CLL, HCL, FL, and MCL render targeted molecular confirmatory testing (eg, BRAF V600E mutation in HCL, t[11;14] in MCL) possible, thus facilitating faster and more effective diagnostics. For example, the diagnostic approach described herein allows for a reliable differential diagnosis between MCL and CLL in virtually every case.…”

“… 22 A particular challenge for the model is the inclusion of DLBCL cases that were missing in a previous publication. 17 CD10 + DLBCL interfered with the correct classification of a fraction of FL and BL cases, whereas a subgroup of CD10 − DLBCL cases were difficult to distinguish from LPL and MZL exclusively using phenotypic data ( Table 4 ). Not surprisingly, DLBCL therefore belonged to the disease categories with the lowest PPV and NPV.…”

UMAP Based Anomaly Detection for Minimal Residual Disease Quantification within Acute Myeloid Leukemia

“…However, these methods can only learn fixed decision boundaries to separate biologically meaningful sub-populations and can not adapt to variations between samples. A way to circumvent this and process whole samples instead of single cells is proposed in [ 44 ]. Here self-organizing maps are employed to obtain a 2D image from a given FCM sample.…”

Translational Applications of Artificial Intelligence and Machine Learning for Diagnostic Pathology in Lymphoid Neoplasms: A Comprehensive and Evolutive Analysis

“…Regarding this, several works have reported the crescent use of AI and ML tools in the diagnosis of hematological diseases [14,15]. Among hematological malignancies, lymphoid neoplasms (LN) constitute one of the most active foci of research in this area, and different AI algorithms have been developed to improve accuracy in lymphoma subtyping [16,17], validation of prognostic models [18], and prediction of chemotherapy response [19,20]. However, a global analysis of the major trends, leading producers, and scientific mapping of AI and ML applications to diagnostic pathology in LN has not yet been undertaken.…”