AI Model Predicts Immunotherapy-Related Pneumonitis Risk from Routine CT Scans Before Treatment

CIPHER AI Model Identifies High-Risk Patients Before Treatment

Researchers at MD Anderson Cancer Center have developed an AI model called CIPHER (Checkpoint-Inhibitor Pneumonitis Hazard EstimatoR) that predicts immunotherapy-related pneumonitis risk by analyzing routine chest CT scans taken before treatment begins1

1

2

2

. The foundation model addresses a critical challenge in cancer care: pneumonitis, a potentially life-threatening side effect that affects approximately 10% of lung cancer patients receiving immunotherapy1

1

. Published in Journal for ImmunoTherapy of Cancer, this research demonstrates how standard medical imaging contains previously unrecognized clues about patient vulnerability to treatment-induced lung inflammation2

2

.

Training on 590,000 CT Image Slices

The AI model was trained using more than 590,000 CT image slices from 2,500 patients with lung cancer1

1

. Rather than learning directly from confirmed pneumonitis cases, CIPHER first learned to recognize subtle imaging patterns within lung tissue, then evaluated whether those patterns could identify patients who later developed the condition after receiving immunotherapy2

2

. This approach allowed the model to detect meaningful signs of lung vulnerability rather than simply identifying future cases. Lead researcher Jia Wu, Ph.D., associate professor of Imaging Physics and Thoracic/Head and Neck Medical Oncology at MD Anderson Cancer Center, explained that the model identified signals associated with future risk using information already present in routine CT scans1

1

.

Achieving 0.83 AUC Across Multiple Datasets

Researchers tested CIPHER using pretreatment CT scans from 347 patients with non-small cell lung cancer treated at MD Anderson Cancer Center and validated the approach using external datasets2

2

. The AI model achieved an area under the curve (AUC) of approximately 0.83 in both cohorts, outperforming conventional clinical-factor models and radiomics approaches1

1

. 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 detects genuine biological vulnerability2

2

. The model's predictions remained significant even after accounting for factors such as age, smoking history, tumor histology and prior thoracic radiation exposure1

1

.

Implications for Personalized Monitoring and Prevention

This AI model could enable personalized monitoring and prevention strategies for lung cancer patients receiving immunotherapy. By identifying high-risk patients before treatment begins, clinicians could implement closer monitoring protocols or consider preventive interventions before serious immunotherapy toxicities develop2

2

. The approach suggests routine imaging contains substantially more information about treatment toxicity than previously recognized, according to Wu1

1

. Future prospective studies will explore whether combining imaging data with other biomarkers can further improve risk prediction and whether similar AI approaches can predict additional immunotherapy-related toxicities1

1

. Researchers also plan to evaluate whether the model performs similarly in other cancer types treated with immunotherapy, potentially expanding its clinical utility beyond lung cancer patients2

2

.