main concepts scientists sections
SimplePopularAdvanced

AI Compass: Which Breast Cancer Patients Really Need Chemotherapy ✎

Original: "Causal multi-modal AI for personalized chemosensitivity prediction"
· Dhruva Biswas, Jeroen Berrevoets, Alec McClean, Linus Bao, Jungkyu Park, Ken G. Zeng, Joseph Cappadona, Cerise Tang, Chuwen Liu, Bartosz Machura, Yin Wu, Valerie Speirs, Hatem Soliman, Rohit Bhargava, Sheheryar Kabraji, Thaer Khoury, David Page, Brian Piening, Carlo Bifulco, Claudia Meurs, Pieter Westenend, Sylvie Chabaud, Jerome Lemonnier, Paul H. Cottu, Florence Dalenc, Fabrice Andre, Frederique Madeleine Penault-Llorca, Thomas Bachelot, Frederick Howard, Francisco J. Esteva, Kevin Kalinsky, Lajos Pusztai, Jan Witowski, Krzysztof J. Geras
arXiv:2609.13567v1 · 2026-09-11 · arXiv non-exclusive · independent analysis · 2 min
Artificial Intelligence
A new model called CTX uses routine tissue images and clinical data to predict personal chemotherapy benefit more accurately than standard genetic tests.

A ship with hundreds of passengers is caught in a storm, and the captain hands out life rings. Some people need a ring, while for others it will only get in the way. Oncology faces the same challenge: after surgery to remove a breast tumor, doctors prescribe chemotherapy, but it is not clear in advance which patients will benefit and which will only get side effects with no gain.

Until now, the main tool has been the Oncotype DX genetic test, which assesses the risk of recurrence but not sensitivity to treatment. It is as if the captain handed out rings based on luggage weight rather than swimming ability. The new CTX model looks at the tumor itself: at its microscopic images, where cell shape and arrangement are visible, and at routine data from the patient's chart, such as age, stage, and receptors.

The model was trained on 9,141 patients from 12 hospitals in 9 countries and tested on 1,994 patients from 5 independent clinics.

The key move is to compare the outcomes of similar patients who did and did not receive chemotherapy. For each woman, the model predicts two scenarios: how the disease will progress with hormonal therapy alone and what will change if chemotherapy is added. The difference between these predictions is the personal benefit, the very life ring that will either keep her afloat or become extra weight.

The results are encouraging. In patients with high predicted benefit, chemotherapy reduced the risk of cancer recurrence by 58%, while in those the model predicted low benefit, there was no effect at all. In a direct comparison on the same women, CTX proved more accurate than Oncotype DX.

If doctors followed the model's advice, chemotherapy could be prescribed 30% less often without worsening outcomes.

The model works on routine data: a standard tissue image and chart notes, without expensive genetic tests. This opens the door to personalized oncology in clinics that lack sophisticated equipment. And the compass that shows which passengers truly need a ring is already ready to come aboard.

Whole graph →

The Falcon model used to analyze histology slides was trained on two billion tissue patches, roughly 40,000 whole slides, comparable to the archive of a major cancer center accumulated over decades.

The idea of predicting an individual's response to treatment echoes the film Gattaca, where genetic analysis determines a person's fate. Here, however, it is not about fatalism but about the ability to choose the optimal therapy based on data.

Concepts
radiotherapy ai planning medical imaging ai chemotherapy tumor imaging characterization tumor segmentation molecular property prediction brain cancer detection personalized medicine cancer drug discovery machine learned interatomic potential radiation therapy quantum chemical calculation solar flare prediction proton therapy photographic body analysis protein language model dna repair