Combining the prognostic values of entropy-based heterogeneity features from 18 F-fluorodeoxyglucose PET and...

The objective of this study is to evaluate the combined prognostic values of 18 F-fluorodeoxyglucose ( 18 F-FDG) PET and computed tomography (CT)-derived entropy-based heterogeneity features from hybrid PET/CT scanner using machine learning in patients with lung adenocarcinoma undergoing curative surgery.

Presurgical 18 F-FDG PET/CT from 131 patients with lung adenocarcinoma were divided into training ( n  = 92) and temporal validation ( n  = 39) cohorts. In the training cohort, we integrated entropy-based heterogeneity features from 18 F-FDG PET/CT for disease-free survival (DFS) prediction using machine learning approach. The predictive value of clinical variables and 18 F-FDG PET/CT-based machine learning for DFS was examined using Cox regression analyses, and independent prognosticators were used to develop the survival prediction model. The model was then tested in the temporal validation cohort.

In the training cohort, 18 F-FDG PET/CT-based machine learning, female sex, and pN status independently predicted DFS. The model, incorporating these predictors significantly predicted DFS in the training (hazard ratio = 1.483, P  < 0.001) and validation cohorts (hazard ratio = 1.753, P  < 0.001). This model outperformed traditional staging system in both cohorts (c-indices = 0.717 vs. 0.621 in training; and 0.728 vs. 0.644 in validation). The model also predicted overall survival in both cohorts (hazard ratio = 1.370, P  < 0.001 in training; hazard ratio = 1.574, P  = 0.017 in validation).

Our preliminary results suggest that integrating prognostic values from 18 F-FDG PET and CT-based heterogeneity features with clinical prognosticators is feasible and may support personalized treatment strategies for patients with resectable lung adenocarcinoma.