AI Finds Two Hidden Features Inside Breast Cancer Tumours That Could Predict Survival

Artificial intelligence has uncovered previously hidden patterns inside breast cancer tumours that could help doctors predict how aggressively a cancer may behave and identify patients at greater risk of poorer outcomes.

Researchers from the University of Southampton have developed a new AI system capable of analysing tiny structures inside tumour cells called centrosomes. The technology can examine hundreds of thousands of cells from tumour samples, revealing differences that are difficult to detect with conventional methods.

The findings could eventually help improve breast cancer prognosis and support more personalised treatment, although further research is needed before the technology can be used routinely in clinical care.

AI Reveals What Is Hidden Inside Tumours

The system, called CenSegNet, was developed as a high-throughput deep-learning framework for analysing centrosomes at both cellular and spatial levels.

Centrosomes are small structures involved in organising microtubules, which help cells divide and maintain their internal structure. Abnormal centrosomes have previously been linked to cancer progression, but studying them across entire tumours has been difficult because of their tiny size and the enormous amount of data involved.

CenSegNet allows researchers to map these structures across large tumour samples and identify patterns that vary between different parts of the same cancer.

Two Hidden Patterns Could Help Predict Outcomes

The research revealed distinct centrosome patterns associated with tumour heterogeneity, disease progression and clinical outcomes.

One important feature was centrosome number, with abnormal increases in centrosome number associated with cancer biology. The researchers were also able to examine centrosome size and spatial organisation, providing a more detailed picture of how these structures differ between cells and regions within a tumour.

These features matter because breast cancer is not a single disease. Cells within the same tumour can behave differently, meaning that examining only a small section of tissue may fail to capture the full biological picture.

By mapping centrosomes throughout tumour samples, the AI system can identify differences between regions and potentially reveal characteristics associated with more aggressive disease.

Why Tumour Heterogeneity Matters

Tumour heterogeneity is one of the major challenges in cancer treatment. A single tumour can contain groups of cells with different characteristics, including differences that influence how quickly the cancer grows, spreads or responds to treatment.

The new approach gives researchers a way to study those differences at a much greater scale.

Instead of manually examining individual structures, CenSegNet can analyse large numbers of cells and create detailed maps showing where particular centrosome patterns occur. This could help scientists better understand how breast cancers evolve and why some tumours behave more aggressively than others.

Potential for More Personalised Cancer Treatment

The researchers say the technology could eventually help clinical professionals identify patients at higher risk and improve predictions about how breast cancer may progress. It could also provide information that supports the development of more targeted therapies by identifying biological features associated with aggressive tumours.

However, the findings do not mean AI can currently predict an individual patient's survival with certainty. CenSegNet remains a research tool, and additional studies will be needed to determine how reliably the identified patterns can be used in clinical decision-making.

The significance of the research lies in what it makes possible: AI is giving scientists a way to examine cancer at a microscopic level and uncover patterns that might otherwise remain invisible.

As researchers continue combining artificial intelligence with digital pathology, these hidden features could eventually become another piece of the puzzle in determining which breast cancers are most dangerous and which treatments are most likely to work.