AI CT Scans May Predict Immunotherapy Pneumonitis

How Can AI Predict Immunotherapy Pneumonitis?

The Checkpoint-Inhibitor Pneumonitis Hazard EstimatoR, known as CIPHER, is a deep-learning model developed by researchers at The University of Texas MD Anderson Cancer Center. It was trained using more than 590,000 CT image slices from 2,500 people with lung cancer, allowing it to learn patterns within lung tissue before researchers evaluated whether those patterns could predict immune checkpoint inhibitor-related pneumonitis.

In the peer-reviewed study published in the Journal for ImmunoTherapy of Cancer, researchers tested the model using pretreatment scans from 347 patients with non-small cell lung cancer and assessed it in an independent external dataset. Its area under the receiver operating characteristic curve was approximately 0.83 in both cohorts, outperforming the clinical-factor and conventional radiomics approaches evaluated by the investigators.

Why Is Immunotherapy-Related Pneumonitis Dangerous?

Immune checkpoint inhibitors help the immune system recognize and attack cancer, but they can also trigger inflammation in healthy organs. Pneumonitis is an uncommon but potentially serious immune-related adverse event involving lung tissue. Warning signs can include a new or worsening cough, shortness of breath, chest discomfort, fever or reduced oxygen levels, although some cases are first detected on imaging.

Diagnosis requires clinical judgment because respiratory infections, pulmonary embolism, tumor progression and other treatment toxicities can produce similar findings. Management depends on severity and may involve pausing immunotherapy, administering corticosteroids and investigating alternative causes, consistent with professional oncology guidance. Patients receiving immunotherapy should promptly report new breathing symptoms rather than waiting for their next appointment.

Could This AI Model Change Lung Cancer Treatment?

Because routine chest CT scans are already part of lung cancer care, CIPHER could potentially add risk information without requiring a new invasive procedure. A validated tool might support more frequent symptom checks, earlier imaging or referral to pulmonary specialists for people identified as higher risk. It could also help researchers select participants for trials of pneumonitis-prevention strategies.

The model is not yet a replacement for oncologists, radiologists or established diagnostic testing. The reported work was retrospective, and performance must be confirmed prospectively across broader populations, cancer types and treatment settings. Researchers also need to determine whether acting on its predictions improves outcomes and whether errors could lead to unnecessary testing, anxiety or changes in effective cancer therapy.