AI tracks thousands of brain tumorcells in living tissue

Researchers at Uppsala University have developed an AI-powered imaging method that can track thousands of individual brain tumorcells as they move through living tissue. The technology, called GlioTrace, combines automated microscopy with machine learning to reveal how glioblastoma cells spread, change their behaviour and respond to cancer treatments. Published in Nature Communications, the study provides new insights into the invasive growth of one of the most aggressive forms of brain cancer.

Glioblastoma is particularly difficult to treat because tumorcells infiltrate surrounding healthy brain tissue, often extending beyond the area that can be surgically removed. These remaining cells can contribute to tumorrecurrence, even after surgery and additional treatment.

Understanding how individual cancer cells migrate through brain tissue is therefore an important research priority. Until recently, scientists largely relied on static microscopic images, which revealed where tumorcells were located but provided limited information about how their behaviour changed over time. GlioTrace offers a different approach by allowing researchers to observe and analyse tumorcell movement continuously over several days.

Tracking thousands of cells with AI

The research team studied glioblastoma cells obtained from six patients. These cells were grown as tumors in mouse brains, which were subsequently cut into thin slices and maintained as living tissue in the laboratory. Using automated microscopy, the researchers captured images at intervals of no more than two hours over periods of up to five days. A machine learning system identified individual tumorcells and tracked their movements. Statistical models were then used to analyse changes in cell behaviour and transitions between different movement patterns.

Unlike imaging techniques that examine relatively small areas of brain tissue, GlioTrace can follow thousands of cells across entire brain slices using standard microscopes. According to Hitesh Mangukiya, a researcher at Uppsala University's Department of Immunology, Genetics and Pathology, this approach makes it possible to study tumorinvasion as a dynamic process rather than relying on isolated observations. The method also allows researchers to compare glioblastoma cells originating from different patients and investigate how their behaviour changes following treatment.

Tumorcells adapt their movement

The analyses revealed considerable differences between the investigated tumors. Some glioblastoma cells primarily migrated along blood vessels, while others spread more diffusely through the surrounding brain tissue. Importantly, individual cells were not restricted to a single movement pattern. The researchers observed that the same tumorcell could change its shape and switch between different migration strategies within just a few days. Some cells moved along blood vessels before changing to independent movement through the tissue.

This flexibility may help explain why glioblastoma is particularly difficult to control. Treatments aimed at limiting one type of movement may not necessarily prevent tumorcells from adopting another strategy. Madeleine Skeppås, a doctoral researcher and co-lead author of the study, emphasised that understanding these transitions is essential for developing treatments capable of limiting tumorinvasion.

Different cell responses

The researchers also used GlioTrace to examine how cancer treatments influence tumorcell migration. Two experimental drug substances reduced cell movement and invasion in the living brain slices. However, their effects varied between patient-derived tumors and between different populations of migrating cells. The team additionally investigated the effects of combined radiation and chemotherapy.

In some tumormodels, this treatment increased the proportion of cells displaying a particular shape. These cells were among the fastest moving and occupied a central position in the transitions between different cellular states. The findings raise the possibility that certain invasive cell populations may persist during treatment. However, the researchers caution that further studies are needed to establish whether these cells are genuinely more resistant to treatment or contribute to tumorrecurrence. The results demonstrate how analysing individual cells over time can reveal biological responses that might remain unnoticed when researchers examine only tumorsize or static tissue samples.

AI also accelerates brain tumordiagnosis

The Uppsala research reflects a broader development in which artificial intelligence is helping scientists extract more detailed information from brain tissue and medical images. AI accelerates brain tumor diagnosis As we previously reported, researchers at the German Cancer Research Centre (DKFZ), Heidelberg University and Heidelberg University Hospital developed an AI system capable of classifying brain tumors using routinely stained microscopic tissue samples.

The system, called Hetairos, was trained and validated using more than 11,000 digitised tissue sections from 9,606 patients across eleven medical centres on four continents. It can distinguish between 102 molecular subtypes of central nervous system tumor. For cases where the system produced a high-confidence classification, accuracy reached approximately 87–88%. In a prospective evaluation involving 210 samples, Hetairos generated classifications within approximately twelve minutes, compared with an average of twelve days for conventional molecular diagnostics.

The two technologies address different aspects of brain cancer research and diagnosis. Hetairos aims to accelerate and support tumorclassification, potentially helping pathologists determine which additional investigations are necessary. GlioTrace, by contrast, investigates how individual glioblastoma cells behave within living tissue and how their movement changes under treatment. Together, these developments illustrate how AI can support both the identification of brain tumorcharacteristics and a deeper understanding of the biological processes that make these cancers difficult to treat. Neither system is intended to replace specialist clinical judgement.

Targeted brain cancer research

A central component of GlioTrace is its computational analysis software, developed in collaboration with researchers at Chalmers University of Technology and the University of Gothenburg. The software processes microscopy images, identifies individual cells and their shapes, and calculates how their movement patterns change over time. The researchers have made the software freely available, enabling other research groups to apply the method. Sven Nelander, professor of integrative cancer research at Uppsala University, believes the technology could help identify mechanisms involved in tumorinvasion and support the development of treatments designed to prevent cancer cells from spreading through the brain.

For now, GlioTrace remains an experimental research method. It has been evaluated using living brain tissue maintained in laboratory conditions, not as a diagnostic tool in patients. Further studies are needed to determine whether the insights gained from tracking individual tumorcells can eventually contribute to more effective treatment strategies. Its immediate value lies in revealing the changing behaviour of thousands of cancer cells, providing researchers with information that conventional imaging approaches have struggled to capture.