Putting the I in AML: Artificial Intelligence and Machine Learning in Acute Myeloid Leukemia

Acute myeloid leukemia (AML) produces more molecular, imaging, and clinical data per patient than any hematologist can hold in mind at once, and each revision of the WHO, ICC, and European LeukemiaNet (ELN) frameworks adds to the load. Artificial intelligence (AI) and machine learning (ML) now reach into every stage of AML care. Deep-learning models read therapy-relevant mutations directly from bone-marrow smears; automated flow-cytometry gating reproduces expert calls in under a minute; and the first AI pathology devices for hematology have cleared regulatory review and entered clinical use. Beyond diagnosis, ML captures the age-dependent weight of individual mutations that categorical ELN scoring misses, drug-response prediction for venetoclax–azacitidine has been validated across multiple external cohorts, and large language models are being tested for tumor-board support and trial matching. The next wave, from clonal-architecture modeling and single-cell foundation models to digital twins and reinforcement learning for adaptive dosing, could move AML management from reactive toward predictive, evolution-aware care. This review departs from existing AI-in-hematology surveys in three ways: we (i) restrict scope to AML and organize the field around clinical decision points rather than technology categories, (ii) grade every tool on a five-tier clinical-readiness level (CRL-AML 1–5), which exposes hundreds of models clustered at CRL-AML 1–2 and none yet in prospective clinical evaluation, and (iii) close with a numbered three-year agenda naming the consortia, datasets, and pragmatic trials needed to carry the field from publication to practice.