How Artificial Intelligence Is Transforming Leukemia Detection and Diagnosis

Artificial intelligence (AI) is rapidly reshaping the landscape of leukemia diagnosis, offering new possibilities for earlier detection, more precise classification, and improved patient care, according to a new review.¹

Traditional methods for diagnosing leukemia rely heavily on expert interpretation of blood smears, bone marrow evaluation, and immunophenotyping, which can be timely, resource-intensive, and difficult to standardize across institutions. As global leukemia incidence continues to vary by region, age, and socioeconomic conditions, researchers continue to explore scalable tools that can reduce diagnostic delays and improve outcomes. In the newly published review, appearing in F1000 Research, researchers have highlighted how advances in medical imaging and machine learning are beginning to fill this critical gap.

“A coherent synthesis of the literature suggests that the future of AI in leukemia lies not in producing isolated classification accuracies but in providing clinically relevant decision support,” explained the researchers. “Automated tools that triage suspicious smears could accelerate confirmatory testing, while lineage-aware algorithms aligned with WHO-5/ICC enable precise subtype identification. When coupled with NCCN-oriented prompts such as recommending TP53 or IGHV testing in CLL, or molecular milestone tracking in CML these systems can bridge the gap between computational detection and therapeutic action.”

This current review synthesized more than 25,000 scientific articles, using a PRISMA-guided approach to analyze both global epidemiological patterns and the application of AI technologies in diagnosing the 4 major leukemia types: acute lymphoblastic leukemia (ALL), acute myeloid leukemia (AML), chronic lymphocytic leukemia (CLL), and chronic myeloid leukemia (CML).

AI-driven image analysis has emerged as one of the most promising innovations in this space. Convolutional neural networks (CNNs), a type of deep learning model designed to interpret visual data,² consistently outperformed traditional methods in identifying leukemic cells in microscopic blood smear images. These tools follow a structured pipeline: image acquisition, preprocessing to enhance clarity, segmentation to isolate target cells, feature extraction to quantify shape and texture, and final classification into normal or malignant cells. This automated workflow reduces the subjectivity and variability inherent in manual review and can dramatically accelerate diagnosis, particularly in low-resource settings or regions where trained hematologists are scarce, explained the researchers.

Several studies highlighted in the review demonstrate notable accuracy. Models using transfer learning with pre-trained networks like AlexNet achieved up to 100% correct classification in controlled datasets, while hybrid systems combining CNN-based feature extraction with machine learning classifiers such as support vector machines (SVMs) reached accuracy levels above 97%.

Other approaches integrated deep autoencoders to handle complex feature patterns, further improving sensitivity and specificity in distinguishing leukemia subtypes. Importantly, these advances extend beyond simple detection and indicate that emerging models can now differentiate among various ALL and AML subtypes, a crucial factor that directly informs treatment selection and prognosis.

Despite these breakthroughs, significant challenges remain. AI systems depend heavily on large, high-quality, annotated datasets—resources that are unevenly distributed across the globe. Differences in staining techniques, imaging hardware, and patient populations introduce variability that limits model generalizability.

Some studies noted performance fluctuations when algorithms were tested across platforms or institutions, emphasizing the need for standardized data and external validation. Ethical and practical barriers also persist, including the integration of AI into clinical workflows, regulatory approval, and the risk of widening inequities if advanced tools are only accessible to well-resourced health systems.

Beyond confirming known risk factors, including genetic syndromes such as Down syndrome, exposure to benzene and ionizing radiation, and viral infections, the analysis underscored wide disparities in access to diagnostic resources around the world. High-income regions like North America, Australia, and Western Europe report higher leukemia incidence but also lower mortality, reflecting the role of early detection and advanced treatment availability. In contrast, developing regions face delayed diagnoses and increased mortality, reinforcing the need for accessible diagnostic solutions.

References

1. Achir A, Debbarh I, Zoubir, et al. Advanced in leukemia detection and classification: a systematic review of AI and image processing techniques. 2025;13:1536. doi:10.12688/f1000research.159318.2

2. Yamashita R, Nishio M, Gian Do RK, Togashi K. Convolutional neural networks: an overview and application in radiology. 2018;9:611-629. doi:10.1007/s13244-018-0639-9