| Summary: | The Myelodysplastic Syndromes (MDS) are a biologically and clinically heterogeneous group of bone marrow haematopoietic cell disorders that result in ineffective haematopoiesis. Unlike most forms of haematological malignancy, the diagnosis of MDS remains heavily reliant on subjective morphological interpretation which can result in inaccurate and missed diagnoses. The use of flow cytometric immunophenotyping offers a potential solution to aid in the diagnosis of MDS, and numerous flow cytometric scoring schemes have been already been proposed and tested. However, most flow cytometric scoring schemes are user-defined, with simple schemes lacking diagnostic sensitivity, whilst the more comprehensive schemes may be unfeasible to implement in a large-scale diagnostic setting. The use of machine learning classifiers offered a more subjective approach to the use of flow cytometric data. Therefore, we have tested a series of classifiers both by combining simple immunophenotypic and demographic features, and by utilising a 2 tube-immunophenotyping panel which contained a large array of numerical and immunophenotypic attributes which had been identified as being abnormal in MDS patients. We have shown that machine learning classifier-based approaches could reproducibly identify patients with definite abnormalities in MDS, and those with normal haematopoietic populations in non-diagnostic, reactive conditions. The classifiers further offered the ability to aid in the triage of patients unlikely to be MDS by providing the basis to a diagnostic confidence score. The application of multiple classifiers also identified a grey-area of MDS patients who were consistently misclassified and who may prove to be challenging to diagnose by flow cytometry, due to an absence of aberrant immunophenotypic features. Finally, we have also shown that a combination of immunophenotyping and targeted gene mutation analysis provides the potential to identify non-diagnostic cases which may progress to MDS. It is in a combination of these two techniques where the future of MDS diagnosis may lie.
|
|---|
