CT-Based Deep Foundation Model for Predicting Immune Checkpoint Inhibitor-Induced Pneumonitis Risk in Lung Cancer.

Area of Science:

• Artificial intelligence in medical imaging
• Deep learning for predictive diagnostics
• Oncology and immunotherapy research

Background:

• Immune checkpoint inhibitors (ICIs) offer advanced cancer treatment but can cause severe immune-related adverse events (irAEs), notably pneumonitis (ICI-P).
• Early identification of patients at high risk for ICI-P is crucial for proactive monitoring and timely intervention to optimize treatment outcomes.

Purpose of the Study:

• To develop and validate a deep learning foundation model, CIPHER, for predicting ICI-P risk using baseline CT scans in lung cancer patients.
• To assess the model's performance against established radiomic approaches.

Main Methods:

• Designed CIPHER, a deep learning model using contrastive learning and a transformer-based masked autoencoder, pre-trained on a large dataset of non-small cell lung cancer (NSCLC) CT scans.
• Fine-tuned and validated CIPHER on internal and external NSCLC cohorts, comparing its predictive accuracy against classical radiomic models.

Main Results:

• CIPHER demonstrated strong performance in predicting ICI-P, with AUCs ranging from 0.77 to 0.85 in internal validation and 0.83 in external validation.
• The model outperformed radiomic models in head-to-head comparisons, showing superior specificity without compromising sensitivity in external validation.

Conclusions:

• CIPHER is the first AI foundation model for immune toxicity prediction, successfully developed and validated for forecasting ICI-P risk from pre-treatment CT scans.
• Prospective validation and integration into clinical workflows could significantly improve patient management and outcomes in immunotherapy.