Review of flow cytometry findings and associated scoring approaches for identifying myelodysplastic syndrome and the future role of machine learning in improving the diagnostic algorithm.

Myelodysplastic syndrome (MDS) is a heterogeneous set of neoplasms that require careful exclusion of potential mimics before diagnosis. In the absence of identified recurrent cytogenetic or molecular genetic alterations, low-grade MDS can be particularly challenging to diagnose. Despite years of accumulating data demonstrating atypical flow cytometry findings associated with MDS, flow cytometry fails to find significant widespread use for diagnosing MDS primarily due to the varied and often subtle altered flow cytometry findings observed. To address this issue, several groups have developed and reported scoring systems to help discriminate MDS from non-MDS using flow cytometry. Our objective in this article is to review the published scoring systems, as well as emerging role of machine learning in MDS diagnosis. Herein, we review many of the recurrent flow cytometric findings and associated reported scoring systems. We also review recent applications of machine learning to MDS and discuss its potential for enabling widespread use of the technology to assist in diagnosing MDS. Several of the published scoring systems are modified versions of or additions to the Ogata scoring system, with the ELN iFS score performing well in comparison studies. New and evolving machine learning approaches have the potential to facilitate improved use of flow cytometry for faster and more accurate MDS detection and have helped identify more useful features, such as erythroid cell SSC. Flow cytometry provides additional information that can help in diagnosis of MDS. Multiple scoring systems now exist that have significant potential to improve the standard approach to diagnosing MDS. Although requiring further development and validation, machine learning methods appear promising as an even more sensitive, specific, and rapid approach to using clinical flow cytometry for identification of MDS than the various prior reported scoring methods. We look forward to furthering improvements in flow cytometry for evaluation of patients with MDS, particularly through the developing use of machine learning approaches and other computational methods.