An exploratory study on whole-lung radiomics features from computed tomography for prognostic prediction in non-small cell lung cancer with concurrent chemoradiotherapy.

Background: Several studies have suggested that lung tissue heterogeneity is associated with the poor prognosis of non-small cell lung cancer (NSCLC). The prognostic value of ipsilateral lung tissue radiomics on predicting outcomes for patients with NSCLC undergoing concurrent chemoradiotherapy (CCRT) remains unclear. This study was conducted to see if ipsilateral whole-lung radiomics would be better at predicting the prognosis of patients with NSCLC CCRT with just the information of the tumor. Methods: We included pre-treatment computed tomography (CT) images that were collected from patients with NSCLC undergoing CCRT from January 1, 2019 to December 31, 2023 for training (n=131) and validation (n=33) sets in this multicenter, retrospective study. Radiomics features of tumor and radiomics whole lung based on feature extraction were derived from the delineation of primary tumor and ipsilateral whole lung (without the primary tumor area) on CT image. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation based on the minimum criteria. Eight machine learning algorithms based on selected features were used to develop the tumor-based [two-dimensional (2D) and three-dimensional (3D)], ipsilateral whole-lung-based, and integrated radiomics models. The model with the best performance was identified using the area under the receiver operating characteristic curve (AUC), which was used to stratify patients into high/low risk groups using Youden-index-derived thresholds for predictive response assessment. Results: Among all models, Extreme Gradient Boosting (XGBoost) demonstrated superior performance. Particularly, whole-lung-based radiomics models yielded higher AUCs than tumor-based approaches, with training set values of 0.936 (2D-Rad), 0.908 (3D-Rad), and 0.782 (Lung-Rad), while validation set results were 0.678, 0.686, and 0.710, respectively. In terms of each model's performance, when combining the two models, the radiomics and lungs had the best performance in the training set and validation set with AUCs of 0.966 and 0.816, respectively. The patients were divided into high-risk groups and low-risk groups according to the threshold value of the combined model. There was a significant difference in progression-free survival (PFS) and overall survival (OS) between the two groups (P<0.05). Conclusions: This study provides the first evidence that ipsilateral whole-lung radiomics has independent prognostic value beyond tumor-focused features in patients with NSCLC undergoing CCRT. Our combined prediction model, based on radiomics, demonstrated the best predictive performance for patients with NSCLC receiving CCRT with different risk levels.