Bytes to blood: Artificial intelligence in leukemia management—A 2025 update.

Abstract

e18557 Background: Artificial intelligence (AI) has been proposed as a tool to aid in the diagnosis, treatment, and monitoring of leukemias given their genetic complexity, subtype heterogeneity, array of treatments, and need for relapse detection. AI has several potential applications in the management of leukemia. First, it can be used to detect leukemia. Using AI to detect subtle nuances in lab values can ensure these deadly cancers are never missed and that complications, such as disseminated intravascular coagulation, are recognized quickly. Second, AI can be used to risk-stratify patients and personalize treatments. Leukemias are among the most genetically complex cancers and, while they have well-characterized risk profiles, tailoring treatments remains a challenge despite our advances. And third, AI can be used for surveillance ( e.g. , minimal/measurable residual disease [MRD] testing) and relapse management. There is a role for AI in predicting risk of relapse and determining when treatment is indicated for a patient who is MRD + , as not all leukemic cells detected can cause relapse. Using AI to tailor treatment plans for patients who ultimately do experience relapse is an especially intriguing potential role for AI. The objective of this narrative review is to determine the current state of AI utilization in the management of leukemia, specifically in diagnosis, risk stratification, treatment planning, and relapse detection. Methods: PubMed was indexed for articles published between January 2022 and November 2025 using the search term "artificial intelligence" AND "leukemia" together. Initial search identified 355 full-text articles. Upon critical appraisal by the authors, 55 articles were deemed relevant to the scope of our work and included in this review. Generative AI was not used in any aspect of the production of this manuscript. The authors have no financial disclosures. Results: The use of AI in the management of leukemia is largely limited to diagnosis, specifically via interpretation of routine clinical data ( e.g. , blood counts, blood smears, metabolic labs) and ancillary testing ( e.g. , flow cytometry, FISH, sequencing). The use of AI in leukemia risk stratification, treatment planning, and relapse detection is largely unexplored. Conclusions: AI has extraordinary potential to improve several aspects of leukemia management. However, it is currently underutilized and primarily focused on initial diagnosis. The real power of AI lies in optimizing treatment intensity to reduce short- & long-term toxicity and the likelihood of relapse, thereby improving patient outcomes. A. P. Runde and S. M. Koo contributed equally.