The potential promise of machine learning in myelodysplastic syndrome.

Abstract

The introduction of artificial intelligence (AI), and in particular machine learning (ML), has revolutionized biomedical research at the clinical level, a trend that also includes hematologic malignancies and myeloid neoplasia (MN). ML encompasses a wide range of applications such as enhanced diagnostics, outcome predictions, decision trees and clustering. Despite several reports in recent years and the achievement of promising results, none of the ML-based pipelines have been directly translated into clinical practice. ML offers the potential to help refine risk stratification and increase accuracy to correctly predict clinical outcomes and disease classification. One of the complications in the clinical utilization of ML is that a large percentage of hematologists have limited familiarity with these tools which can cause skepticism. Concerns have also been raised by patients that are worried about privacy issues, reliability of the outcomes, and loss of human interaction. In this review, we aim to pinpoint the main mechanisms and applications of ML, as well as application in MN and Myelodysplastic Syndrome, highlighting strengths and limitations, and addressing the potential promise in clinical implementation of ML-pipelines.