Radiomics and Deep Learning Prediction of Immunotherapy-Induced Pneumonitis From Computed Tomography

Area of Science:

• Radiology and Oncology
• Artificial Intelligence in Medicine

Background:

• Immune checkpoint inhibitors (ICI) offer significant cancer treatment benefits but can cause severe side effects like pneumonitis (PN).
• Predicting PN risk is essential for safe and effective long-term ICI therapy and patient stratification in clinical trials.

Purpose of the Study:

• To develop and evaluate imaging-based models for predicting PN in patients undergoing ICI therapy.
• To compare the performance of radiomics and deep learning models in forecasting PN risk.

Main Methods:

• A cohort of 671 cancer patients with available pre-treatment chest CT scans from 3,351 ICI-treated patients was analyzed.
• Three predictive models were developed: one using radiomics features, one using a convolutional neural network (CNN) on raw CT data, and a combined model.

Main Results:

• The CNN-only model achieved a higher Area Under the Curve (AUC) of 0.819 compared to the radiomics-only model (AUC 0.747).
• The combined radiomics and deep learning model showed a marginal improvement in AUC (0.829) but not statistically significant over the CNN-only model.
• The CNN model demonstrated improved sensitivity (0.743) and comparable positive predictive value (0.244) for PN prediction.

Conclusions:

• Deep learning models, particularly CNNs, show superior utility in predicting PN compared to traditional radiomics.
• These findings suggest a potential for improved patient management and risk stratification in ICI therapy and clinical trials.