Predicting PD-L1 status in NSCLC patients using deep learning radiomics based on CT images
Jiameng Lu1,2, Xinyi Liu1, Xiaoqing Ji3
1Department of Respiratory, The First Affiliated Hospital of Shandong First Medical University & Shandong Provincial Qianfoshan Hospital, Shandong Institute of Respiratory Diseases, Shandong Institute of Anesthesia and Respiratory Critical Medicine, 16766 Jingshilu, Lixia, Jinan, 250014, Shandong, People's Republic of China.
Scientific Reports
|April 11, 2025
Summary
Deep learning radiomics (DLR) effectively predicts programmed death-ligand 1 (PD-L1) expression in non-small cell lung cancer (NSCLC) using CT scans. Integrating DLR with clinical data further improves prediction accuracy, aiding personalized NSCLC treatment strategies.
Area of Science:
- Radiomics and Medical Imaging
- Oncology
- Artificial Intelligence in Medicine
Background:
- Programmed death-ligand 1 (PD-L1) expression is a key biomarker for immunotherapy in non-small cell lung cancer (NSCLC).
- Accurate prediction of PD-L1 status is crucial for guiding treatment decisions in NSCLC patients.
- Current methods for assessing PD-L1 expression can be invasive or time-consuming.
Purpose of the Study:
- To develop and evaluate a deep learning radiomics (DLR) approach for predicting PD-L1 expression in NSCLC patients.
- To assess the performance of DLR models using CT imaging features.
- To investigate the added value of integrating DLR with clinical data for enhanced PD-L1 prediction.
Main Methods:
- Utilized CT images from 352 NSCLC patients with known PD-L1 status.
- Employed semi-automated segmentation for tumor regions of interest (ROI).
- Extracted deep learning features using Residual Network 50, followed by LASSO for feature selection. Developed and compared DLR models and an integrated DLR-clinical data model using ROC analysis.
Main Results:
- The DLR model achieved an AUC of 0.85 for predicting PD-L1 status from CT images.
- The integrated model combining DLR and clinical data demonstrated superior performance with an AUC of 0.91.
- Both models showed significant sensitivity and specificity in PD-L1 status prediction.
Conclusions:
- Deep learning radiomics (DLR) shows significant potential for non-invasively predicting PD-L1 status in NSCLC.
- The integration of DLR with clinical data offers enhanced predictive accuracy.
- This DLR approach can assist in clinical decision-making and optimizing personalized treatment strategies for NSCLC.
