Combination of CT derived radiomic features and lymphovascular invasion status to predict disease recurrence following trimodality therapy in non-small cell lung cancer.

Authors:

Abstract
E24314 Background: Currently there are no clinically validated biomarkers to identify which locally advanced non-small cell lung cancer (NSCLC) patients (pts) will benefit from trimodality therapy (i.e. neoadjuvant chemoradiation (NAT) followed by surgery, TMT). In this study, we evaluate radiomic (i.e. computer extracted imaging) features of tumor phenotype as potential predictors of disease recurrence following TMT. Methods: 123 pts with stage III NSCLC treated with TMT were included. Of these, 33 pts with either distant metastasis at presentation and/or those without baseline pre-NAT CT scans were excluded. Lung tumors were retrospectively annotated by an expert reader. 1542 radiomic features (textural and shape) were extracted from intra and peritumoral regions. multivariate Cox regression analysis was performed to evaluate radiomic and clinico-pathologic factors predictive of recurrence post-surgery. Results: Pathological response was achieved in 36 (40%) pts and the rest 54 (60%) were labeled non-responders. 90 pts were randomly allocated to equally sized (n = 45) training and validation sets. The recurrence rate after surgery was 42.2% (n = 38) and the mean duration of recurrence was 20.28 ± 16.43 months. Multivariate Cox regression model revealed that a combination of two radiomic features (intratumoral and peritumoral Laplacian Laws features that capture micro-gradients and patterns of heterogeneous enhancement) from a baseline non-contrast CT scan as well as lymphovascular invasion (LVI) signature status was predictive of time to recurrence after TMT (p-value = 0.0283, 0.0129, 0.0454 respectively). Three-fold cross validation was used for training a Support Vector Machine (SVM) classifier with three features on the training set and subsequently evaluated on the validation set. The classifier successfully distinguished pts based on time to recurrence post-TMT, which was statistically significant (log-rank test, p-value < 0.014). Conclusions: Combination of LVI status and texture features extracted from within and around the tumor on baseline CT images was predictive of disease recurrence in NSCLC pts treated with TMT.