Routine chest scans before immunotherapy can help clinicians spot lung injury risk early, according to new research.
RT’s Three Key Takeaways:
- Early Risk Detection: An artificial intelligence foundation model accurately identifies lung cancer patients at heightened risk of immunotherapy-induced pneumonitis by analyzing standard pretreatment chest computed tomography scans.
- Superior Predictive Accuracy: The model achieved an area under the curve of approximately 0.83 in both internal and external cohorts, outperforming conventional clinical-factor and imaging approaches.
- Clinical Monitoring Implications: Patients identified as high risk developed lung inflammation sooner after initiating therapy, providing respiratory and oncology teams with an objective pathway for personalized surveillance and intervention strategies.
Researchers at The University of Texas MD Anderson Cancer Center have developed an artificial intelligence (Ai) model that identifies patients with lung cancer who are at increased risk of developing a serious immunotherapy-related side effect before treatment begins, according to a study published in Journal for ImmunoTherapy of Cancer.
The study indicates that standard medical imaging contains subtle clues regarding patient vulnerability to pneumonitis, a potentially life-threatening lung inflammation that affects approximately 10% of lung cancer patients receiving immunotherapy. By analyzing routine chest computed tomography (CT) scans captured prior to starting therapy, the investigators identified underlying imaging patterns linked to future risk, surpassing current methods that rely on subjective visual review and clinical variables.
The investigation was led by Jia Wu, PhD, associate professor of imaging physics and thoracic/head and neck medical oncology at UT MD Anderson, alongside co-senior authors Ajay Sheshadri, MD, associate professor of pulmonary medicine, and Mehmet Altan, MD, 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,” said Wu, associate professor of imaging physics and thoracic/head and neck medical oncology at UT MD Anderson, in a news release. “Our model was able to identify signals associated with future risk using information that already exists in routine CT scans.”
Training and Validating the CIPHER Model
The investigative team developed the Checkpoint-Inhibitor Pneumonitis Hazard EstimatoR (CIPHER), an Ai foundation model trained on more than 590,000 CT image slices obtained from 2,500 lung cancer patients. Instead of training solely on confirmed pneumonitis cases, the model learned baseline tissue characteristics across the lung and then evaluated whether those patterns could predict which individuals developed lung inflammation following immunotherapy.
Investigators tested CIPHER on pretreatment CT scans from 347 patients with non-small cell lung cancer (NSCLC) managed at UT MD Anderson, validating the tool on an independent external dataset. The system produced an area under the curve (AUC) of approximately 0.83 in both cohorts, exceeding the performance of conventional radiomics approaches and clinical risk factors.
According to the study, CIPHER maintained stable performance despite variations across patient cohorts, imaging protocols, and CT equipment. Patients identified as high risk experienced pneumonitis earlier in their therapeutic courses, indicating the platform detects genuine lung vulnerability rather than simply marking eventual cases. The risk predictions also maintained statistical significance after controlling for age, smoking status, tumor histology, and previous thoracic radiation exposure.
“What makes this approach particularly interesting is that it was not designed to look for pneumonitis itself,” said Wu, in a news release. “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.”
Next Steps for Clinical Integration
Before CIPHER can be embedded into routine clinical healthcare workflows, researchers noted that prospective validation across larger, diverse populations is necessary, according to the news release. The team plans to evaluate the model’s accuracy across other cancer types managed with immunotherapy regimens.
Subsequent studies are slated to investigate whether integrating CT image data with additional biomarkers improves risk stratification, and whether comparable Ai frameworks can identify other immunotherapy-related adverse events, according to the researchers. Such developments may ultimately enable oncology and respiratory care teams to better identify candidates for prevention trials, tailor monitoring schedules, and advance understanding of treatment-induced toxicities.
