Abstract
Manuscript version:
Introduction: Flow cytometry is essential for the diagnosis of acute myeloid leukemia (AML), but conventional analysis is labor-intensive, operator-dependent, and increasingly complex. Machine learning (ML) may support automated analysis; however, its readiness for clinical implementation remains unclear. This scoping review mapped ML approaches applied to flow cytometry for AML-related diagnosis.
Methods: PubMed was searched for English-language primary studies applying artificial intelligence, ML, or deep learning to human flow cytometry data for AML diagnosis or classification. Eligible studies reported diagnostic performance metrics. Study characteristics, preprocessing, data representation, model architecture, validation strategy, performance metrics, interpretability, and implementation considerations were descriptively synthesized according to PRISMA-ScR.
Results: Fifty-four records were identified, and eight studies published between 2010 and 2025 were included. Clinical tasks included AML detection, differentiation from non-neoplastic conditions, acute leukemia lineage classification, prediction of selected molecular abnormalities, and classification of AML remission states. Approaches ranged from artificial neural networks and traditional supervised classifiers using statistically derived features to hybrid pipelines combining unsupervised representation learning with supervised classification and end-to-end deep learning models using event-level data. Although most studies reported high diagnostic performance, comparisons were limited by heterogeneity in patient populations, instruments, antibody panels, preprocessing methods, clinical tasks, and reported metrics. Most models underwent internal validation only. One study performed independent cross-institutional validation, while another evaluated prospective clinical implementation and post-deployment monitoring.
Conclusion: ML-assisted flow cytometry shows potential for AML screening, classification, triage, and workflow support, but current evidence does not support autonomous diagnosis. Clinical translation will require standardized inputs, transparent preprocessing, interpretable outputs, external and prospective validation, continuous quality monitoring, and integration into locally validated laboratory workflows.
Document Type
Article
Publication Date
8-2026
Recommended Citation
Dela Pena, Katrina Jezzela M. MD, MLS (ASCP), "Machine Learning in Flow Cytometry for Acute Myeloid Leukemia Diagnosis: A Scoping Review" (2026). Other Specialties. 62.
https://researchrepository.parkviewhealth.org/other/62