Use of PET derived features to predict mutational status in lung adenocarcinomas

Objectives: The treatment of pulmonary adenocarcinomas is nowadays driven by the presence/absence of some genetic abnormalities defining distinct molecular sub-groups of patients with various outcomes. The aim of our study was to determine the ability of a large number of parameters derived from initial pre-treatment 18FDG PET images to predict the mutational status.

Methods: 498 patients with pulmonary adenocarcinoma who underwent 18FDG PET/CT before treatment were retrospectively included. All primary tumors (tumor volume larger than 4 cm³) were identified visually on PET images and were automatically segmented with the FLAB algorithm. Derived radiomic parameters were confronted with EGFR, KRAS, BRAF mutational status, and PDL1 expression. ALK translocation and ROS1 fusion were also considered. The performance of each PET derived feature was evaluated using ROC methodology.

Results: A total of 120 features were extracted from PET images. Several parameters were efficient in predicting EGFR, PDL1, ROS and BRAF status with an individual AUC reaching 0.86. The best predictive parameters were different depending on the gene studied. None of the evaluated parameters could predict KRAS and ALK status (AUC <0.65). Routinely used SUVmax had no significant predictive value for any of the studied genetic profiles. in addition, textural parameters performed better than any tumor volume related parameters.

Conclusions: Our results confirm that PET derived radiomics could be a powerful approach to identify specific molecular subgroups of pulmonary adenocarcinoma and should be considered to guide patient management. A model based on a selection of the best correlating parameters to each genetic profile will be tested and results will be presented.