AI tool reveals hidden patterns inside breast tumours

An open-source AI tool has analysed more than 330,000 cellular structures in breast tumours, revealing two forms of abnormality that were previously treated as one.

Breast cancers can vary within the same tumour, making it difficult to understand which changes signal aggressive disease or may respond to targeted treatment.

University of Southampton researchers developed CenSegNet to map centrosomes, structures that organise cell division, at single-cell resolution. Working with University Hospital Southampton, they studied 911 tumour specimens from 127 patients. The system distinguished cells with too many centrosomes from those with enlarged ones and found that the patterns occupied different tumour regions. Enlarged centrosomes were associated with more aggressive features, while fewer enlarged structures in the tumour core were linked with better overall survival.

The result shows how AI can expose biological patterns at a scale that manual analysis cannot easily match. These are associations, however, and CenSegNet is not ready to guide routine patient care.

Researchers will now combine the maps with genetic and molecular data to test whether they can become reliable treatment biomarkers.