Preselection of robust radiomic features does not improve outcome modelling in non-small cell lung cancer based on clinical routine FDG-PET imaging

Abstract Background Radiomics is a promising tool for identifying imaging-based biomarkers. Radiomics-based models are often trained on single-institution datasets; however, multi-centre imaging datasets preferred external generalizability owing to the influence of inter-institutional scanning differences and acquisition settings. The study aim was determine value preselection robust radiomic features in routine clinical positron emission tomography (PET) images predict outcomes locally advanced non-small cell lung cancer (NSCLC). Methods A total 1404 primary tumour were extracted from pre-treatment [ 18 F]fluorodeoxyglucose (FDG)-PET scans stage IIIA/N2 or IIIB NSCLC patients using training cohort ( n = 79; prospective Swiss randomized phase III trial SAKK 16/00; 16 centres) an internal validation 31; single centre). Robustness studies investigating delineation variation, attenuation correction motion performed (intraclass correlation coefficient threshold > 0.9). Two 12-/24-month event-free survival (EFS) overall (OS) logistic regression standardized imaging: (1) with alone (2) all available features. Models then validated fivefold cross-validation, separate single-centre dataset. Model performance assessed area under receiver operating characteristic curve (AUC). Results identified 179 stable (13%), 25% 3D versus 4D acquisition, 31% 78% delineation. Univariable analysis found no significant predicting EFS 12-month OS p 0.076). Prognostic without well (AUC 0.73) 0.74). Patient stratification into two risk groups based 0.02) cohorts 0.03). Conclusions PET-based radiomics model standardized, dataset successfully established good performance. Prediction feature unsuccessful, indicating need protocol.