Artificial intelligence is beginning to change how leukemia is diagnosed, particularly by turning blood and bone-marrow slides into digital data that algorithms can analyze. Traditional diagnosis combines morphology with flow cytometry, cytogenetics and molecular profiling, and the article argues that AI’s immediate role is not to replace hematopathologists but to make visual information more measurable and reproducible. AI systems can detect and classify blood and bone-marrow cells, quantify morphological features and identify patterns associated with acute leukemia.
One promising area is automated cell recognition and classification. Deep-learning systems have been studied for detecting abnormal blasts, distinguishing normal from leukemic cells, and differentiating types of acute leukemia. Such tools could screen thousands of cells, highlight suspicious populations and help laboratories handle large specimen volumes more consistently. In time-sensitive diseases such as acute promyelocytic leukemia, AI could potentially flag suspicious cases for urgent expert review and confirmatory molecular testing, although it would not replace definitive testing.
However, the article stresses that high AI accuracy does not automatically mean accurate leukemia diagnosis. A model may perform extremely well at identifying individual cells in a carefully controlled dataset but struggle with real-world specimens containing staining differences, damaged or overlapping cells, unusual morphologies and rare leukemia subtypes. Differences in scanners, laboratory procedures and patient populations can also reduce performance. External validation across institutions and prospective testing in real clinical workflows are therefore essential before these systems can be trusted routinely.
The longer-term opportunity is multimodal AI, combining digital morphology with flow cytometry, cytogenetics, sequencing, blood counts and clinical information. This could help hematopathologists integrate large amounts of diagnostic evidence more efficiently rather than relying on AI for a single classification task. The article ultimately presents AI as an assistant that could automate repetitive screening and allow specialists to focus on difficult cases requiring judgment, interpretation and additional testing. The key challenge is moving from impressive retrospective models to reliable, validated clinical tools that improve diagnosis without creating automation bias or replacing human oversight.
