Session Item

Several studies have shown that lung heterogeneity is associated with overall survival (OS) in lung cancer. However, the predictive value of whole lung-based radiomics for OS in non-small cell lung cancer (NSCLC) is unknown. The purpose of this study is to investigate the predictive value of whole lung-based and tumor-based radiomics for OS in locally advanced-NSCLC treated with curative radiotherapy.

A total of 349 patients were enrolled in the study, with 292 patients in training set, 57 patients from the same hospital as independent test-set-1, and 47 patients from a multi-institutional prospective clinical trial data set (RTOG0617) as test-set-2. Tumor-based radiomics features and total lung-based radiomics features were extracted from planning CT images and radiomics features were selected by least absolute shrinkage embedded with Cox proportional hazards model with 5-fold cross-internal validation, with 1000 bootstrap samples. Radiomics prognostic scores (RS) were calculated by Cox regression based on selected features. Five models based on clinical factors, a tumor RS, and a lung RS separately and their combinations were constructed. The Harrell concordance index (C-index), calibration curves were used to evaluate the discrimination and calibration performance. Patients were stratified into high and low risk groups, and a log-rank test was performed.

Figure 1: Workflow of the project. Abbreviations: L-RM: COX regression model based on lung signature; TL-RM: COX regression model based on tumor signature and lung signature; CTL-RM: COX regression model based on tumor signature, lung signature and clinical factors; C-index: Harrell concordance index.

The discrimination ability of lung- and tumor-based radiomics model was similar in terms of C-index, 0.69 vs 0.68 in training set, 0.69 vs 0.66 in test-set-1, 0.63 vs 0.69 in test-set-2. The model that combined clinical factors, tumor and lung RS achieved the best performance (C-index, 0.78 in training set, 0.80 in test-set-1, 0.81 in test-set-2). The calibration curve showed good agreement between predicted and actual values. Patients were well stratified, with statistical significance for all datasets.

Figure 2: Kaplan-Meier survival curves based on the CTL-RM model in (A) training set, (B) test-set-1, (C) test-set-2. Abbreviations: CTL-RM: COX regression model based on tumor radiomics signature, lung radiomics signature and clinical factors.

Lung- and tumor-based radiomics features have the potential to predict OS in NSCLC. The combination of clinical and radiomics predictors can achieve optimal performance.