Prognostic and predictive value of radiomics features at MRI in nasopharyngeal carcinoma

• DOI: 10.1007/s12672-021-00460-3
• Publication Date: 2021-12-17T18:21:00Z
• Authors: Dan Bao; Yanfeng Zhao; Zhou Liu; Hongxia Zhong; Yayuan Geng; Meng Lin; Lin Li; Xinming Zhao; Dehong Luo
• Keywords: LASSO Cox regression analysis; Disease progression; 03 medical and health sciences; Radiomics; Magnetic resonance imaging; 0302 clinical medicine; Research; Nasopharyngeal carcinoma; Neoplasms. Tumors. Oncology. Including cancer and carcinogens; RC254-282; 3. Good health

Abstract Purpose To explore the value of MRI-based radiomics features in predicting risk disease progression for nasopharyngeal carcinoma (NPC). Methods 199 patients confirmed with NPC were retrospectively included and then divided into training validation set using a hold-out (159: 40). Discriminative radiomic selected Wilcoxon signed-rank test from tumors normal masticatory muscles 37 patients. LASSO Cox regression Pearson correlation analysis applied to further confirm differential expression set. Using multiple model, we built feature-based classifier, Rad-Score. The prognostic predictive performance Rad-Score was validated cohort illustrated all Results We identified 1832 differentially expressed between tissue. based on one feature: CET1-w_wavelet.LLH_GLDM_Dependence-Entropy. showed satisfactory predict an area under curve (AUC) 0.604, 0.732, 0.626 training, validation, combined (all included) respectively. improved stratification, progression-free survival significantly different these groups every (p = 0.044 or p < 0.01). Combining clinical features, higher AUC achieved prediction 3-year (PFS) (AUC, 0.78) 5-year PFS 0.73), although there no statistical difference. Conclusion Rad-Score, proven useful pretreatment prognosis potential stratification NPC.