Identifying 18F-FDG PET-metabolic radiomic signature for lung adenocarcinoma prognosis via the leveraging of prognostic transcriptomic module

Abstract

Background: Imaging with 18F-fluorodeoxyglucose positron emission tomography (18F-FDG PET), which identifies molecular and metabolic abnormalities within tumor cells, could support prognostic assessment of lung adenocarcinoma (LUAD). We aimed to develop a radiomic signature with the aid of a transcriptomic module for individualized clinical prognostic assessment of LUAD patients. Methods: Using a gene expression profile consisting of 334 stage I-IIIA LUAD patients, prognostic-related gene coexpression modules were constructed via a weighted correlation network analysis algorithm. The robustness and prognostic performance of the coexpression modules were then tested across 2 gene expression datasets totaling 331 patients. Finally, using a discovery dataset with matched transcriptomic and 18F-FDG PET radiomic data of 15 patients and multiple linear regression analysis, we developed a PET-metabolic radiomic signature that had optimal correlation with the expression of a robust prognostic module. Results: We selected a superior coexpression module for LUAD prognosis in which the genes were significantly enriched in important biological processes associated with tumors (e.g., cell cycle, DNA replication and p53 signaling pathway). The prognostic performance of the module for overall survival (OS) and recurrence-free survival (RFS) was validated in 2 independent gene expression datasets (log-rank P<0.05). Through the leveraging of this prognostic coexpression module, a radiomic signature consisting of 3 PET features associated with metabolic processes was developed in the discovery dataset. The radiomic signature was significantly associated with patients' OS and RFS in an independent PET dataset consisting of 72 LUAD patients (OS: Log-rank P=0.0006; RFS: Log-rank P=0.0013). Multivariate Cox analysis demonstrated that the radiomic signature was an independent prognostic factor for OS and RFS. Furthermore, the novel proposed radiomic nomograms for OS and RFS had significantly better performance (concordance indices) than did the clinicopathological nomograms. Conclusions: The radiomic signature, which reflects biological processes in tumors (e.g., cell cycle and p53 signaling pathway), could noninvasively identify LUAD patients with poor prognosis who should receive postoperative adjuvant treatment. The signature is suitable for clinical application and could be robustly applied at an individual level across multicenter cohorts.