A team from Stanford Medicine has developed and validated an artificial intelligence model capable of detecting rare secondary cancers that arise years after radiation therapy for an initial malignancy. The research, published in Nature Medicine on January 15, 2026, addresses one of the most challenging aspects of survivorship care: the late emergence of radiation-induced tumours, which often have poor prognosis due to delayed diagnosis.

Radiation therapy remains a cornerstone of treatment for many childhood cancers, lymphomas, breast cancer, and head-neck cancers, but it carries a small but serious long-term risk of causing secondary malignancies such as sarcomas, thyroid cancer, breast cancer in the radiation field, and certain brain tumours. These cancers can appear 5–30 years after treatment, and conventional surveillance relying on periodic physical exams and symptom-triggered imaging frequently misses early, asymptomatic lesions.

The Stanford team, led by radiation oncologist and data scientist Dr. Susan Hiniker and bioinformatician Dr. Erqi Pollom, trained a deep learning model on a large retrospective cohort of over 4,000 survivors followed for decades at Stanford and collaborating institutions. The AI analyses serial imaging (MRI, CT, and PET scans), radiation treatment plans, and longitudinal clinical data to identify subtle textural changes, volume anomalies, and metabolic shifts that are highly predictive of emerging secondary cancers.

In validation testing, the model demonstrated excellent performance: it flagged secondary malignancies with a sensitivity of 89% and specificity of 94%, often detecting lesions 12–36 months before they were diagnosed clinically. Importantly, the AI reduced false positives compared to traditional radiologist review, minimising unnecessary biopsies and patient anxiety.

The tool is now being piloted at the Stanford Cancer Institute’s Late Effects Clinic, where survivors of childhood and young adult cancers are followed lifelong. Plans are underway to refine the model for broader deployment, potentially integrating it into national survivorship guidelines and making it available to other cancer centres through a cloud-based platform.

Dr. Susan Hiniker, lead author of the study, said: “Survivors of radiation therapy live with a lifelong risk of secondary cancers that are often aggressive and hard to detect early. This AI tool gives us a powerful new way to monitor these patients proactively, catching problems when they are most treatable sometimes years before symptoms or conventional imaging would raise concern.”

The work is part of Stanford’s broader effort to use AI for precision surveillance in oncology, particularly for high-risk survivor populations. Collaborators included faculty from radiation oncology, radiology, biomedical informatics, and paediatric oncology, along with support from the Stanford Cancer Institute and the National Cancer Institute.

As the number of cancer survivors continues to grow globally, tools like this could transform long-term follow-up care shifting it from reactive to predictive and personalised, ultimately improving survival and quality of life for those cured of one cancer but at risk of another.