Breast cancer AI preprint uses inferred gene expression to predict therapy response
Published in Artificial Intelligence, Journals, Personalized medicine.
Published October 5, 2026. Source date: October 2, 2026.
A new arXiv preprint reports an AI approach for estimating whether breast cancer will show a pathological complete response to treatment given before surgery. The model first infers gene-expression information from tissue images, then combines those estimates with clinical variables.
Researchers trained the first stage on data from 8,742 patients spanning 32 cancer types. The response-prediction model was developed with 1,080 patients across five cohorts and evaluated with 1,412 patients across nine cohorts. Its pooled AUROC was 0.79, with a 95% confidence interval of 0.73–0.85. AUROC measures discrimination rather than the percentage of patients correctly diagnosed.
The findings suggest a potential route to biologically informed prediction from biopsy images. This is a preprint: the reported evaluation does not establish prospective clinical benefit, justify replacing genomic testing, or demonstrate regulatory approval.