Pathologic-radiomic mixed model predicts survival in operated non-small cell lung cancer

Abstract

Abstract Purpose The objective of our retrospective study was to assess the prognostic value of texture analysis and histopathological features in surgically resected lung cancer patients. Materials and methods In total, 70 patients with lung cancer stage IC to IIIA were included in this study. Tumor sections were morphologically evaluated on the basis of tumoral infiltrating lymphocytes, stromal density and tumor necrosis. CT texture analysis was performed using commercially available software (TexRAD) on unfiltered and filtered images with four spatial scale filters. Relevant textural features were selected using Least Absolute Shrinkage and Selection Operator (LASSO) with internal cross-validation. Hazard ratios (HR) were calculated using an overall textural coefficient splitting the sample at an optimal cut-point. Prognostic significance of pathological variables was evaluated with Cox models. The comparison between the performance of the textural-based model, the pathological-based model and the combined model was evaluated by Brier score and cross-validated c-index. Results Entropy spatial scale filter (SSF) = 2 was related to overall survival (OS). Patients with different textural scores had significant OS differences (p = 0.011, HR = 2.29). Trends were noted for pathological features and patients were classified into two groups with different prognoses (p = 0.007, HR = 2.8). Tumors with higher Hounsfield units (HU) or unfiltered mean value of positive pixels (mpp) were associated with histopathological features (HU, p = 0.008 and mpp, p = 0.005). The combination of textural and pathological features gave three different prognostic groups and the combined textural plus pathological model was the most informative and most accurate (c-index 0.708). Conclusions Textural and pathological tumor analyses are both prognostic and complementary in risk stratification. If these results are confirmed in future studies, they could form the basis for modifying treatment decisions for patients. Advances in knowledge CT texture analysis could help in prognostic stratification of surgically operated lung cancer patients and is complementary to histopathological analysis.