AI could diagnose blood disorders and infections in minutes, not days

IRR researchers have developed an AI model capable of identifying different blood cells and detecting early signs of infection using black-and-white microscope images. The approach could make diagnostics quicker and more accessible, particularly in settings without specialised lab infrastructure.

Moving beyond slow, expensive lab tests

Currently, diagnosing and monitoring blood disorders and infections typically involves trained specialists examining stained blood samples under a microscope or using expensive techniques such as flow cytometry.

This process is time consuming and requires special equipment and stains to make the cells visible in colour.

A recent pair of studies, led by the Bachmann group, show how an AI-based imaging platform can both identify different blood cell types and detect infection.

AI can distinguish between different blood cells

In the first study, the team trained an AI model to recognise red blood cells, white blood cells (immune cells) and platelets from thousands of simple grayscale microscope images. The model also used technology originally designed to detect everyday objects in photographs to build on (from the School of Informatics, University of Edinburgh).

The AI could correctly identify cells by their shape and structure about 85% of the time, performing particularly well on red blood cells. The AI’s performance was like that of some high-end lab machines used by expert analysers and was achieved without chemical stains.

These findings could help make blood analysis faster, cheaper, and more accessible, especially in places that don’t have full labs. It could also help automate parts of hospital diagnostics.

Dr Alex Hunt IRR researcher (Bachmann lab) and paper’s first author

Although there are existing methods for doing blood cell identification and counting, the use of computer vision and AI methods has opened up the possibility of doing this cheaper and faster. The potential here is exciting.

Prof Robert Fisher Chair in Computer Vision, School of Informatics

Detecting infection before symptoms escalate

In the second study, researchers explored whether the same AI model could detect when white blood cells respond to infection.

When bacteria invade the body, white blood cells quickly change shape as they activate to fight off infection. These subtle shape changes can act as early warning signs but are difficult to detect quickly with current methods.

The AI system was trained to recognise both the type of white blood cell and whether it was in an activated state. It proved especially effective at identifying activation in neutrophils and lymphocytes.

Three separate white blood cells as unactivated and activated
The AI system was trained to recognise both the type of white blood cell (lymphocyte, monocyte or neutrophil) and whether it was in an activated state (responding to infection) or not

The researchers then introduced small amounts of E. coli bacteria into blood samples, at levels similar to real-life bloodstream infections. The AI successfully detected immune cell activation at very low bacterial levels and within 30 minutes of incubation. Once images were captured, analysis took less than five minutes.

This is comparable to the sensitivity of current, standard blood culture methods, but far faster.

Faster, cheaper, more accessible diagnostics

Together, the two studies demonstrate the potential of a stain-free AI platform for blood analysis. By relying on cell shape rather than chemical markers, the system eliminates the need for costly reagents and reduces dependence on specialist staff.

All that’s required is a basic microscope and trained AI software, allowing for faster diagnostics in hospitals, clinics, and in low-resource settings. This new platform could reshape how and where blood tests could be performed.

While further development and validation is needed before clinical use, there is a lot of potential in this research. The same technology could, for example, be adapted to detect viral or fungal infections, identify blood cancers such as leukaemia, or diagnose conditions such as anaemia.

For time-critical illnesses such as sepsis, the ability to detect infection in minutes rather than days will be life-changing.

Prof Till Bachmann IRR Group Leader and paper’s corresponding author

Tags

CIR
Publication date