The predictive value of 18F-FDG PET/CT radiomics for pleural invasion in non-small cell lung cancer

Abstract

Objective

This study aims to develop and validate a PET/CT radiomics fusion model for preoperative predicting pleural invasion (PI) in non-small cell lung cancer (NSCLC) patients.

Methods

Data from Center A were divided into a training set (n = 260) and an internal validation set (n = 111), while data from Center B (n = 99) served as the external validation set. Radiomic features were extracted using PyRadiomics. Six feature screening methods and 12 machine learning methods were used to build clinical, PET/CT imaging, and radiomics fusion models. The best-performing model was selected based on accuracy, sensitivity, specificity, and area under the curve (AUC). A nomogram was created using logistic regression with clinical, PET/CT features, and Rad_score.

Results

The PET/CT radiomics fusion model exhibited superior predictive performance. In the internal validation set, it achieved an accuracy of 0.90, sensitivity of 0.88, specificity of 0.92, and AUC of 0.95 (95% CI 0.91–0.99). These metrics were significantly higher than those of the PET/CT imaging model (accuracy 0.83, sensitivity 0.83, specificity 0.82, AUC 0.85) and clinical model (accuracy 0.65, sensitivity 0.70, specificity 0.59, AUC 0.78). In the external validation set, the model demonstrated an accuracy of 0.81, sensitivity of 0.81, specificity of 0.81, and AUC of 0.85 (95% CI 0.77–0.94), outperforming the PET/CT imaging model (accuracy 0.76, sensitivity 0.75, specificity 0.77, AUC 0.80) and clinical model (accuracy 0.68, sensitivity 0.67, specificity 0.68, AUC 0.76). The nomogram showed excellent calibration, with a C index of 0.98 in the test set, 0.95 in the internal validation set, and 0.91 in the external validation set.

Conclusion

The PET/CT radiomics fusion model significantly improves PI prediction accuracy in NSCLC.

Critical relevance statement:Pleural invasion is a critical prognostic factor in lung cancer and a challenge for preoperative CT evaluation. PET/CT radiomics fusion model has the highest predictive value in predicting PI of lung cancer.