1258P - Does FDG PET-based radiomics have an added value for prediction of overall survival in non-small cell lung cancer?

Background

Machine-learning and radiomics are promising approaches to improve the clinical management of NSCLC. However the additive value of FDG-PET based radiomic compared to clinical and standard imaging variables is less investigated. We aimed to assess the prognostic role of combined use of machine-learning and radiomics in NSCLC patients.

Methods

320 NSCLC patients underwent PET/CT with FDG at initial staging. Forty-nine predictors, including 43 textural features extracted from PET studies, SUVmax,MTV, TLG, TNM tumor stage, age and gender were analyzed. A least absolute shrinkage and selection operator (LASSO) regression was used to select features with highest predictive value. Cox's multivariate regression analysis was used to calculate individual risk scores. We divided patients into high-risk and low-risk groups using the median risk score of each prediction model. Kaplan-Meier and log-rank test were used to compare overall survival (OS) of high- and low-risk patients.

Results

Five variables including one textural feature (NGTDM_Coarseness), SUVmax and tumor stage (stage II, stage III, stage IV) showed highest predictive value at LASSO regression analysis. These variables were used to create a LASSO model (tumor stage, SUVmax and NGTDM_Coarseness) which was compared to a reference model (tumor stage, SUVmax). After a median follow-up time of 35 months, 61.6% of patients were deceased. The LASSO model classified 70% deceased and 81% alive as high- and low-risk patients respectively (Odd Ratio(OR)=[9.9, Confidence Interval(CI)=5.8,17.1]). The reference model classified 70% deceased and 82% alive as high- and low-risk patients respectively (OR=[10.7, CI=6.2,18.7]). Kaplan-Meier curves showed that both the reference model and the LASSO model significantly predicted shorter OS in the high-risk group, without any significant model difference: median OS in the high-risk group was 15 months (95%CI, 11-18) for both models while median OS was not reached in the low-risk group for both models.

Conclusions

These data indicate that radiomics does not significantly improve the prediction of outcome based on disease stage and SUVmax. Morover, the SUVmax used with the stage improves the accuracy of the stage alone based prediction.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Andrea Ciarmiello.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.