Predicting Immunotherapy-Induced Pneumonitis Based on Chest CT and Non-Imaging Data

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Immune checkpoint inhibitors (ICIs) improve survival in non-small cell lung cancer but can cause pneumonitis.
• Accurate prediction of ICI-related pneumonitis is crucial for patient management.

Purpose of the Study:

• To develop and evaluate a multi-modality deep learning approach for predicting ICI-pneumonitis.
• To assess the efficacy of combining clinical, radiomic, and deep learning features for prediction.

Main Methods:

• Utilized multi-modal data: clinical records and pre-treatment lung CT scans.
• Extracted deep learning features (vision transformer), radiomic features, and clinical data.
• Compared ten machine learning algorithms for prediction accuracy.

Main Results:

• The multi-modality approach combining all three feature types yielded the best performance.
• Achieved a prediction accuracy of 0.823 and an AUC of 0.895.
• Demonstrated superior prediction compared to single-modality approaches.

Conclusions:

• Multi-modal data integration significantly enhances the prediction of ICI-pneumonitis.
• Machine learning algorithms can accurately identify patients at high risk for this adverse event.
• This approach supports early identification and proactive management of ICI-pneumonitis.