Learning a Cytometric Deep Phenotype Embedding for Automatic Hematological Malignancies Classification.

Area of Science:

• Hematology
• Computational Biology
• Medical Diagnostics

Background:

• Accurate minimal residual disease (MRD) detection is crucial for acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS) prognosis.
• Current flow cytometry (FC) methods for MRD assessment are time-consuming and require expert manual interpretation, posing a bottleneck in clinical practice.

Purpose of the Study:

• To develop an automated algorithm for classifying MRD in AML and MDS using deep learning.
• To improve the efficiency and objectivity of MRD detection in hematological malignancies.

Main Methods:

• A deep phenotype representation learning algorithm was developed using a large dataset of over 2000 patient FC samples.
• The method employs a cell-level autoencoder and specimen-level latent Fisher-scoring vectorization to create a cytometric deep embedding.

Main Results:

• The automated algorithm achieved an average AUC of 0.943 across four hematological malignancy classification tasks.
• Analysis indicated that utilizing only half of the standard FC markers could still yield high classification accuracies.

Conclusions:

• The developed deep learning algorithm offers an efficient and accurate automated solution for MRD classification in AML and MDS.
• This approach has the potential to streamline FC data analysis, reduce inter-observer variability, and advance the treatment of hematological diseases.