The clinical application of artificial intelligence in hematology is undergoing a structural transition. For several years, computational oncology focused heavily on high-cost, low-throughput modalities such as next-generation sequencing and spatial transcriptomics. While these technologies remain vital for academic characterization, they do not address the operational bottlenecks of daily clinical practice. This month’s data demonstrates a clear pivot toward opportunistic screening: leveraging the massive, underutilized streams of routine clinical data—specifically complete blood count (CBC) parameters, raw scattergrams, and standard flow cytometry—to flag hematologic malignancies and rare diseases long before clinical suspicion is triggered.
By embedding interpretable machine learning models directly into existing laboratory workflows, clinical teams can convert routine diagnostic tests into active screening mechanisms. This shift minimizes diagnostic delays, optimizes referral pathways, and provides clinicians with real-time decision support without requiring additional blood draws or expensive specialized assays. As health systems face persistent staffing shortages and rising diagnostic volumes, these pragmatic, high-volume AI applications represent the true operational frontier of digital hematology.
Notable papers
• An interpretable artificial intelligence model for real-time leukemia screening via routine blood tests across multicenter cohorts (npj Digital Medicine)
Researchers validated LeukoAlert, an AI framework that analyzed 446,558 routine complete blood count records from 203,284 individuals across seven institutions, converting standard hematology parameters into a real-time opportunistic screening tool for leukemia and its subtypes.
Brand take: Genuinely useful clinically, as it utilizes existing, low-cost laboratory data to catch aggressive malignancies early without adding to the clinician’s ordering burden.
• AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings (npj Digital Medicine)
This study deployed an AI algorithm across 14 healthcare organizations to screen 1,307,140 patients, successfully identifying 356 high-risk individuals and confirming 11 new cases of paroxysmal nocturnal hemoglobinuria (PNH), a rare disease notorious for diagnostic delays exceeding five years.
Brand take: Genuinely useful clinically, proving that background EHR screening algorithms can solve the needle-in-a-haystack problem for ultra-rare hematologic conditions.
• Self-explaining artificial intelligence for the classification of B cell non-Hodgkin lymphoma: A diagnostic decision support study (PLoS Medicine)
This paper developed a self-explaining AI model to interpret high-dimensional multiparameter flow cytometry data, providing interpretable diagnostic support for B-cell non-Hodgkin lymphoma classification even when working with limited data for rare lymphoma subtypes.
Brand take: Underrated, because its focus on explainability directly addresses the ‘black box’ objection that prevents most flow cytometry AI from achieving clinical adoption.
• Development and validation of a novel multimodal deep neural network model based on CBC digit parameters and scattergrams for rapid hematolymphoid malignancy classification: a multicenter cohort study (BMC Medicine)
This multicenter study validated a multimodal deep learning model that combines standard numerical CBC parameters with raw instrument scattergrams to rapidly classify hematolymphoid malignancies, distinguishing them from reactive states.
Brand take: Genuinely useful clinically, as utilizing raw instrument scattergrams extracts rich, previously discarded diagnostic signals from standard laboratory hardware.
• Advances in digital pathology and artificial intelligence in the diagnosis of myeloid neoplasms (Human Pathology)
This review synthesizes recent developments in digital pathology and AI applied to peripheral blood smears, bone marrow aspirates, and flow cytometry for diagnosing myeloid neoplasms like AML and MDS.
Brand take: Overhyped, because while the image analysis performance is impressive in research settings, the physical workflow bottleneck of digitizing bone marrow biopsies remains a massive barrier to real-world clinical utility.
Products, deals & funding
• Massive Bio & GELL Partnership
Massive Bio partnered with the Latin American Study Group on Lymphoproliferative Disorders (GELL) to deploy its AI-driven clinical trial matching platform across Latin America and the United States, aiming to expand precision oncology and hematology research access.
Brand take: Genuinely useful clinically, as it directly addresses the historical underrepresentation of diverse patient populations in hematologic clinical trials.
• BMS & NVIDIA AI Collaboration
Bristol Myers Squibb expanded its compute infrastructure by deploying NVIDIA’s advanced DGX Vera Rubin systems to scale proprietary AI models and accelerate drug discovery pipelines across hematology and oncology.
Brand take: Overhyped, as massive compute upgrades represent standard corporate infrastructure scaling rather than a direct breakthrough in hematology therapeutic design.
• Sysmex America XR-Series Showcase
Sysmex America announced plans to showcase its next-generation XR-Series hematology solution at the ADLM 2026 Clinical Lab Expo, highlighting its integrated 3D scattergram technology and advanced analytical capabilities.
Brand take: Genuinely useful clinically, as hardware-level integration of advanced analytics is the only way to scale AI-driven hematology screening to high-volume clinical laboratories.
• DKFZ AI System for Blood Cancers
The German Cancer Research Center (DKFZ) presented a new AI system designed to analyze multi-omic data and provide personalized treatment recommendations for complex blood cancers.
Brand take: Underrated, because personalized treatment selection in heterogeneous blood cancers remains highly subjective, and structured decision support is desperately needed.
Regulatory & clinical adoption
The regulatory landscape for hematology AI is shifting toward validating algorithms that run passively in the background of laboratory information systems (LIS) and electronic health records (EHR). The real-world deployment of the PNH screening algorithm across 14 healthcare organizations in Poland, as reported in npj Digital Medicine, demonstrates that clinical adoption no longer requires active clinician interaction at the point of care. Instead, these systems operate as safety nets, flagging high-risk patients for confirmatory testing (such as flow cytometry) based on routine lab trends. Regulatory bodies are increasingly evaluating these ‘opportunistic screening’ tools under frameworks that assess both diagnostic accuracy and the downstream clinical utility of automated alerts, ensuring they do not trigger excessive, unnecessary secondary testing.
Trends & what to watch
The dominant trend in hematology AI is the transition from specialized, high-cost genomic assays to opportunistic screening using existing, high-volume data streams. The success of frameworks like LeukoAlert and multimodal CBC-scattergram models indicates that the next 1-3 months will likely see diagnostic hardware manufacturers increasingly embedding AI models directly into laboratory analyzers. Clinicians should watch for partnerships between diagnostic giants and AI software developers aiming to deliver pre-validated, on-analyzer screening tools that flag suspected leukemia or myelodysplastic syndromes at the moment of sample processing.
Concurrently, explainability has become a non-negotiable requirement for clinical adoption. As demonstrated by the PLoS Medicine study on B-cell non-Hodgkin lymphoma, models that provide clear, interpretable visual or parametric rationales for their classifications are gaining traction over black-box neural networks. For product teams, this means that diagnostic performance alone is no longer sufficient; clinical decision support tools must actively teach the clinician why a specific risk flag was raised, referencing established clinical guidelines and morphologic features.
Bottom line
The future of hematology AI lies in converting routine, high-volume diagnostic tests like CBCs into intelligent, real-time screening networks for hematologic malignancies and rare diseases.