Identification of PD-L1 Expression in Resectable NSCLC using Interpretable Machine Learning Model Based on Spectral...

Area of Science:

• Radiology and Oncology
• Artificial Intelligence in Medicine
• Cancer Biomarkers

Background:

• Programmed death ligand-1 (PD-L1) expression is crucial for predicting treatment response in non-small cell lung cancer (NSCLC).
• Accurate pre-treatment assessment of PD-L1 expression is challenging in resectable NSCLC.

Purpose of the Study:

• To evaluate the efficacy of a machine learning model integrated with spectral computed tomography (CT) for predicting PD-L1 expression in resectable NSCLC.
• To identify key spectral CT features that contribute to PD-L1 expression prediction.

Main Methods:

• Retrospective analysis of 131 resectable NSCLC patients who underwent spectral CT.
• Development and comparison of eight machine learning models using clinical-imaging and spectral CT quantitative parameters.
• Variable selection via logistic regression and LASSO regression; model performance assessed using AUC, sensitivity, specificity, accuracy, F1 score, and decision curve analysis (DCA).

Main Results:

• The extreme gradient boosting (XGBoost) model achieved the highest performance in the test cohort with an AUC of 0.887.
• Key predictors included cavitation, ground-glass opacity, and spectral CT parameters (CT40keV, CT70keV) at the venous phase.
• The XGBoost model demonstrated favorable clinical utility and high prediction accuracy via DCA and calibration curves.

Conclusions:

• Machine learning models incorporating spectral CT quantitative parameters and imaging features show significant potential for predicting PD-L1 expression in resectable NSCLC.
• This approach may aid in non-invasive assessment of PD-L1 status, guiding personalized treatment decisions.