Researchers at Mass General Brigham and the Dana-Farber Cancer Institute have developed and validated a noninvasive, AI-based tool that can predict disease progression in patients with oropharyngeal cancer. The innovation supports more personalized treatment decisions by identifying which patients are likely to benefit from intensified therapy, and which may safely receive less aggressive care.
The AI model focuses on predicting extranodal extension (ENE), a key prognostic factor in head and neck cancer. ENE occurs when cancer spreads beyond the lymph node into surrounding tissue and is associated with poorer outcomes. Until now, ENE could only be confirmed after surgical removal and pathological examination of lymph nodes.
“Our tool helps determine which patients may need multiple or intensified interventions, such as immunotherapy or additional chemotherapy, and which patients could be candidates for treatment de-escalation,” says senior author Benjamin Kann, MD, of the Artificial Intelligence in Medicine (AIM) Program at Mass General Brigham and radiation oncologist at Dana-Farber and Brigham and Women's Hospital. The research was recently published in the Journal of Clinical Oncology.
Addressing a clinical dilemma
Oropharyngeal cancer treatments, including surgery, radiotherapy and chemotherapy, are often effective but can be highly burdensome, with long-term side effects that affect swallowing, speech and quality of life. Clinicians therefore face a delicate balance: treating aggressively enough to control disease, while avoiding overtreatment where possible.
Accurate risk stratification is essential, yet current staging systems provide limited insight into which patients are most at risk of disease spread. ENE is a strong indicator, but its reliance on invasive diagnostics has restricted its use in upfront treatment planning.
Imaging, AI and prognosis
To close this gap, the research team developed an AI algorithm that analyzes routine CT imaging data to estimate the number of lymph nodes affected by ENE, before treatment begins. This represents a significant advance, as the number of ENE-positive nodes correlates with prognosis and potential benefit from intensified therapy.
The model was trained and tested using CT scans from 1,733 patients with oropharyngeal carcinoma. Results showed that the AI tool could reliably predict uncontrolled cancer spread and poorer survival outcomes. When combined with existing clinical risk predictors, the AI-driven assessment significantly improved patient-level risk stratification.
“The ability to predict the number of lymph nodes with ENE is entirely new,” Kann explains. “It provides a powerful prognostic biomarker that could enhance current staging systems and support more tailored treatment planning.”
Data-driven precision oncology
The study highlights how AI-enabled imaging analysis can unlock clinically relevant information hidden in standard diagnostic scans. By transforming routinely collected data into actionable insights, the tool fits squarely within the broader shift toward precision oncology and value-based care.
Looking ahead, the researchers see potential for the AI model to guide enrollment in clinical trials, support shared decision-making and reduce unnecessary treatment-related toxicity. As digital tools become increasingly embedded in oncology workflows, innovations like this illustrate how AI can augment, not replace, clinical expertise, enabling more informed and individualized care for cancer patients.
Lung cancer AI-tool evaluated
Earlier this year British and Dutch researchers evaluated a South Korean AI tool to assess its ability to support lung cancer screening by analyzing low-dose CT scans. Using data from the UK Lung Cancer Screening (UKLS) trial, the study showed that the AI system can reliably identify scans without clinically relevant lung nodules, which make up the majority of cases. This significantly reduces the workload for radiologists, who currently face capacity shortages that hinder large-scale screening programmes. Importantly, all confirmed lung cancer cases were correctly flagged by the AI for further review, ensuring no diagnoses were missed.
The findings suggest that AI could help make population-based lung cancer screening more efficient, affordable and scalable. Researchers emphasize that high-quality imaging data and long-term follow-up were crucial for validation. The study represents a methodological milestone and supports the potential integration of AI into future lung cancer screening programmes.
