Room: Stars at Night Ballroom 2-3
Purpose: To investigate whether prognostic prediction of non-small cell lung cancer (NSCLC) patients who underwent stereotactic body radiotherapy (SBRT) was improved by using not only clinical factors but also radiomic features extracted from breath-hold diagnostic CT images.
Methods: One-hundred seventy-eight primary NSCLC patients (188 tumors) who underwent SBRT at our institution were enrolled in this study. Clinical endpoints were cancer-specific death (CSD), local recurrence (LR), and distant metastasis (DM). For each patient, 1316 radiomic features were extracted from their diagnostic CT images before SBRT. Clinical factors such as age, histology, and biologically-effective dose were also employed for the following analysis. Statistical analysis was performed using Fine and Gray proportional hazard model. Then, three prognostic models were built: a clinical factor-based model (clinical model), a radiomic feature-based model (radiomic model), and a model combining clinical factors and radiomic features (combined model). All cohorts were dichotomized into low- or high-risk groups based on their median risk scores and difference of cumulative incidences were evaluated. To assess the robustness of built models, all cohorts were partitioned into training and test datasets and area under curves (AUCs) in test dataset were calculated.
Results: No prognostic radiomic features were found for LR. For DM, the cumulative incidences of low- or high-risk groups obtained by the combined model yielded more significant difference than those by the other models (p < 0.05). Median AUCs of CSD and DM were 0.621 and 0.608, 0.650 and 0.571, and 0.681 and 0.638 for the clinical, radiomic, and combined model, respectively. Significant improvements of AUCs for CSD and DM were observed for the combined model (p < 0.05).
Conclusion: A prognostic prediction of CSD and DM for SBRT-treated NSCLC patients would be improved by combining radiomic features with clinical factors compared with only clinical factors.
Not Applicable / None Entered.