AI model uses routine imaging to identify patients at risk for serious treatment-induced lung inflammation

UT MD Anderson Research News September 24, 2026

• Immunotherapy-induced lung inflammation is a potentially life-threatening condition that occurs in about 10% of lung cancer patients receiving treatment

• An artificial intelligence model outperformed conventional approaches in predicting patients at elevated risk  

• The model demonstrated consistent performance even on external datasets that included variance in patient populations, types of CT scanners and imaging protocols

• The tool could help identify patients who may benefit from closer monitoring or intervention before serious side effects develop

Researchers at The University of Texas MD Anderson Cancer Center have developed an artificial intelligence (AI) model that can identify lung cancer patients at increased risk of developing a serious immunotherapy-related side effect before treatment begins, offering a possible path toward more personalized monitoring and prevention strategies.

The findings indicate that standard medical imaging may contain clues about a patient’s susceptibility to pneumonitis, a potentially life-threatening form of lung inflammation that occurs in about 10% of lung cancer patients receiving immunotherapy. By analyzing routine chest CT scans obtained before treatment, the researchers identified imaging patterns associated with future risk. This method outpaced current approaches, which rely on subjective imaging analysis and clinical risk factors that do not fully capture underlying vulnerability.

The study, published in Journal for ImmunoTherapy of Cancer, was led by Jia Wu, Ph.D., associate professor of Imaging Physics and Thoracic/Head and Neck Medical Oncology and an affiliate member of UT MD Anderson’s Institute for Data Science in Oncology; co-senior authors Ajay Sheshadri, M.D., associate professor of Pulmonary Medicine; and Mehmet Altan, M.D., associate professor of Thoracic/Head and Neck Medical Oncology.

“Pneumonitis remains one of the most challenging complications of immunotherapy because it can be difficult to predict before symptoms appear,” Wu said. “Our model was able to identify signals associated with future risk using information that already exists in routine CT scans.”

How did the AI model identify patients at higher risk?

Researchers developed the Checkpoint-Inhibitor Pneumonitis Hazard EstimatoR (CIPHER), an AI foundation model trained using more than 590,000 CT image slices from 2,500 patients with lung cancer. Rather than learning directly from confirmed pneumonitis cases, the model first learned to recognize patterns within lung tissue and then evaluated whether those patterns could identify patients who later developed pneumonitis after receiving immunotherapy.

Researchers tested CIPHER using pretreatment CT scans from 347 patients with non-small cell lung cancer (NSCLC) treated at UT MD Anderson and validated the approach using an independent external dataset. The model achieved an area under the curve (AUC) of approximately 0.83 in both cohorts, meaning its predictive power outperformed conventional clinical-factor models and radiomics approaches.

The model maintained strong performance despite differences in patient populations, CT scanners and imaging protocols. Patients classified as high-risk also tended to develop pneumonitis sooner after starting immunotherapy, indicating the model may be detecting meaningful signs of lung vulnerability rather than simply identifying future cases. The model’s predictions remained significant even after accounting for factors such as age, smoking history, tumor histology and prior thoracic radiation exposure.

“What makes this approach particularly interesting is that it was not designed to look for pneumonitis itself,” Wu said. “Instead, the model learned patterns within lung tissue and identified subtle abnormalities associated with future risk. That suggests routine imaging may contain much more information about treatment toxicity than we previously recognized.”

What’s next for this approach?

The findings demonstrate strong predictive performance. Additional prospective studies involving larger and more diverse patient populations will be needed to determine whether the approach can be integrated into clinical workflows. Researchers also plan to evaluate whether the model performs similarly in other cancer types treated with immunotherapy.

Future studies may explore whether combining imaging data with other biomarkers can further improve risk prediction and whether similar AI approaches can predict additional immunotherapy-related toxicities. These tools could help identify candidates for prevention studies, guide monitoring strategies and improve understanding of how treatment-related side effects develop.

***

This research was supported by the National Institutes of Health, the Cancer Prevention and Research Institute of Texas (CPRIT) and UT MD Anderson institutional funding. For a full list of collaborating authors, disclosures and funding sources, see the full paper in Journal for ImmunoTherapy of Cancer.