AI in a hematology analyzer? How HORIBA is reimagining early sepsis detection

Artificial intelligence (AI) is starting to reshape healthcare, but its influence is not just in high-tech imaging or robotic surgeries. One of the most exciting applications of AI is emerging from a less glamorous but critically important area: hematology. The early detection of sepsis, systemic inflammatory response syndrome (SIRS), and septic shock, conditions that can escalate quickly and often prove fatal, may now be achievable with greater speed and accuracy thanks to AI-powered hematology analyzers like HORIBA’s Yumizen H2500 and Yumizen H1500 series.

Sepsis remains one of the leading causes of preventable hospital deaths. Globally, one in five hospital fatalities is linked to sepsis, and the condition is diagnosed in over 47 million people each year. What makes sepsis particularly deadly is the speed at which it progresses. “If there is even one hour delay in treatment, the risk of death increases by 4% to 9%,” says Shubham Rastogi, Medical Field Leader at HORIBA. “That is huge. This is why it is so important to diagnose sepsis at the right time.”

A new role for hematology in critical care

HORIBA’s Yumizen H2500 and H1500 hematology analyzers, typically used for complete blood counts (CBC) and white blood cell differential analysis, are being reimagined as frontline tools in emergency care through the integration of AI algorithms. These algorithms were developed on the data acquired in collaboration with clinical researchers from Christian Medical College in Vellore, India and the Bergonié Institute in Bordeaux, France. The team used retrospective CBC data from over 680 patients between 2018 and 2023 to train and validate the model.

Unlike traditional diagnostic tools like procalcitonin or C-reactive protein (CRP), which are often analyzed in isolation, the AI system processes multiple blood parameters simultaneously to detect patterns that indicate sepsis or its precursors. Rastogi explains this using an analogy, “We named this project the ‘Elephant project.’ If you put an elephant in a dark room and send five people in with torches, one sees the trunk, another sees the leg, another the ear. Everyone describes the elephant differently. That’s how sepsis is diagnosed today, everyone sees only part of it. Our project aimed to deliver consistent indicators by processing multiple parameters simultaneously, allowing us to illuminate wider parts.”

Instead of relying on single-analyte markers, HORIBA’s AI model analyzes cell morphology and blood characteristics using a proprietary LMNE matrix derived from the Yumizen platform’s Double Hydrodynamic Sequential System (DHSS). This unique technology captures detailed cell information through a combination of impedance and optical extinction measurements, offering a much richer dataset for AI to process.

Internal Society of Laboratory Hematology (ISLH) 2025, Halifax, Canada accepted this study which was authored by Dr. Sukesh C Nair, Dr. Francoise Durrieu and Shubham Rastogi, titled: AI-driven human–machine interface for early detection and differentiation of sepsis, SIRS, and septic shock with severity assessment on hematology analyzers.

How does the AI work?

The model uses an unsupervised AI generative method (GeodAIsics’ GML approach) to identify clusters and patterns in the complex biomedical data such as CBC data, rather than relying on pre-labeled training sets. This allows it to differentiate between SIRS, sepsis, and septic shock based on complex interrelationships among several blood parameters. As Rastogi notes, “This feature is something which is differentiating, and helping us to detect more accurately. We are able to differentiate even within SIRS, sepsis, and septic shock.”

According to the team’s research, the AI model on the HORIBA analyzer has achieved a sensitivity of 91% and specificity of 86% on preliminary datasets, with an area under the curve (AUC) of 0.89. This performance outpaces several FDA-authorized AI-based models, which are typically built on immunopanel data rather than hematology. “Other FDA-approved AI models just see sepsis,” says Rastogi. “Our model was able to differentiate sepsis and septic shock, which is a necessity for any hospital.”

This advancement has significant implications for emergency departments and intensive care units, where time is critical. “Emergency is when you need the fastest way of knowing what the problem is,” Rastogi says. “Our technology will help to diagnose early and fast compared to conventional methods.”

Immature granulocytes and large platelet fraction

The AI model is just one piece of the diagnostic potential offered by the Yumizen analyzers. Two other key features, immature granulocyte (IMG) detection and large platelet fraction (LPF) analysis, further enhance the platform’s clinical value. Those are accepted by internationally renowned ISLH 2022, Bologna Italy and Euromedlab authored by Dr. R.K Bhola, S.C Nair, C.Fudaly, S. Rastogi.

IMGs are early-stage white blood cells that typically do not appear in adult peripheral blood unless there is significant immune system activation, such as an infection, trauma, or cancer. Traditional methods for detecting these cells are labor-intensive and unreliable at low counts. HORIBA’s Yumizen analyzers, using DHSS and optical extinction technology, can automatically and accurately identify and quantify IMGs during a routine CBC without additional reagents or procedures.

IMG levels have been linked to disease severity in multiple conditions. For example, elevated IMG counts can indicate severe bacterial infections in pediatric patients, or help predict complications in acute appendicitis and acute respiratory distress syndrome (ARDS). In neonatal ICUs, IMG levels have shown promise as early indicators of sepsis, potentially reducing reliance on invasive and expensive diagnostic tests.

Similarly, the large platelet fraction (LPF), an indicator of the presence of large, immature platelets, can offer insight into bone marrow activity and platelet regeneration. Traditional platelet markers like mean platelet volume (MPV) and platelet distribution width (PDW) suffer from accuracy issues due to interference from fragmented red blood cells. LPF analysis, on the other hand, uses impedance and optical extinction to accurately distinguish large platelets from artifacts, without requiring dye-based staining.

High LPF levels are often seen in conditions with high platelet turnover, such as immune thrombocytopenia or cardiovascular diseases, while low LPF may indicate bone marrow suppression due to chemotherapy or failure. This parameter can also guide transfusion decisions or monitor responses to antiplatelet therapy, making it a valuable addition to any hematology panel.

Learn more about immature granulocyte and large platelet fraction and on the Yumizen H2500/H1500 series hematology analyzers.

Real-time risk assessment

HORIBA’s innovation does not stop at detection. Its AI models have been designed to include a ‘sepsis suspicion flag’ directly on the Yumizen H2500 and H1500 analyzers. This real-time flag alerts clinicians to patients at potential risk of sepsis, enabling faster decisions.

“This is a true innovation for the patients and healthcare professionals,” Rastogi emphasizes. “One out of five deaths are happening globally because of sepsis, and there is no sure-shot option available in the fastest way of diagnosis. So, I think this does contribute to global care.”

HORIBA’s innovation does not stop at detection. Its AI models have been designed to include a ‘sepsis suspicion flag’ directly on the Yumizen H2500 and H1500 analyzers. This real-time flag alerts clinicians to patients at potential risk of sepsis, enabling faster decisions

“This is a true innovation for the patients and healthcare professionals,” Rastogi emphasizes. “One out of five deaths are happening globally because of sepsis, and there is no sure-shot option available in the fastest way of diagnosis. So, I think this does contribute to global care.”

While some elements of HORIBA’s roadmap for AI expansion remain confidential, Rastogi confirms that the company has a “very bright, futuristic plan” for further integration of AI into its diagnostic platforms. “AI in healthcare is still under-used,” he says. “But in the coming future, AI will be new routine. It will not only help diagnosis, but it will also help monitor and treat, at the fastest level.”

Bridging the gap between technology and access

One notable strength of the Yumizen H series is its potential to bring advanced diagnostics to settings with fewer resources. Because the AI run on standard CBC data, the technology can be used in primary healthcare centers and local clinics, not just large urban hospitals. This is critical in regions like Asia and Africa where sepsis incidence is higher than the global range but diagnostic infrastructure is limited.

As Rastogi points out, there has long been a mismatch between where sepsis is most prevalent and where sepsis is most studied. “The present sepsis cases are more in Asia and Africa, but the studies have been mostly in the US and Europe,” he says. “So, our objective was to include both geography and include more to balance this ratio, where the cases are present, and where we have studied. That’s how India and France sites were brought together.”

With its combination of AI capabilities, powerful hematological insights, and accessibility, the Yumizen H2500/H1500 series may well represent a turning point in how we approach early diagnosis in emergency, critical care and patient care.