← Azimuth Daily
Capturing data…
Capturing data…
WED, 4 NOV · 79 ITEMS
"AI-based MANTIS model personalises prognosis for childhood brain tumours" — Nature Genetics · Science & Medicine
Researchers have published a machine-learning model called that generates an individual for children with brain tumours, rather than relying on broad risk groups defined by tumour type or stage. The pitch is treatment personalisation: if the model can rank a child's likely outcome more sharply, clinicians could dial therapy up for high-risk patients and spare low-risk ones from the most toxic regimens. The paper is a methods-and-validation report in Nature Genetics, so the claim is about model performance on the cohorts studied, not yet about changed outcomes in the clinic. The specific input data (molecular/omics profiles, clinical variables, imaging) and the tumour types covered are set out in the paper itself rather than the headline.
This is the primary source: a peer-reviewed Nature Genetics paper (doi:10.1038/s41588-026-02815-4), so the claim is exactly what the authors report and carries the journal's review bar. The thing to watch is validation — prognostic models typically look strongest on the internal or single-institution cohorts they were trained on, and the real test is external, multi-site replication and prospective use.
Genuinely new — the paper was posted within the last day, and there is no recirculation year attached.