Clinical-radiomic features predict survival in patients with extranodal nasal-type natural killer/T cell lymphoma

Purpose To investigate the value of MRI-based radiomic features integrated with clinical indicators for survival prediction in patients with extranodal natural killer/T-cell lymphoma, nasal-type (ENKTL). Materials and methods One-hundred and sixty-five patients with ENKTL who underwent pretreatment MRI were enrolled. Patients were randomly divided into training ( n = 115) and validation ( n = 50) sets. A radiomic signature (R-signature) was generated using the least absolute shrinkage and selection operator regression. Kaplan–Meier analysis and univariate Cox proportional hazards model were used to determine the association of the R-signature and clinical variables with overall survival (OS) and progression-free survival (PFS). Clinical models and combined clinical-R-signature models were constructed by multivariable Cox regression analysis, respectively. Results The R-signature achieved C-index of 0.666 and 0.684 (training set) and 0.679 and 0.691 (test set) for the prediction of OS and PFS, respectively. For both OS and PFS prediction, the C-index was comparable between the R-signature and clinical model both in the training cohort (OS: C-index = 0.666 vs. 0.719, p = 0.284; PFS: C-index = 0.684 vs. 0.725, p = 0.439) and the validation cohort (OS: C-index = 0.679 vs. 0.665, p = 0.878; PFS: C-index = 0.691vs.0.668, p = 0.803), respectively. The combined clinical-R-signature models achieved better predictive performance than the R-signature in the training cohort (OS: C-index = 0.741vs.0.666, p = 0.032; PFS: C-index = 0.762 vs. 0.684 p = 0.020), respectively. The differences did not reach statistical significance in the validation cohort ( p > 0.2). Conclusion The radiomic signature extracted from baseline MRI can predict outcomes of patients with ENKTL, and the combination of MRI radiomic signature and clinical predictors may further improve the predictive performance in patients with ENKTL.