Encrypted login | home

Program Information

MRI Based Radiomics Signature, a Quantitative Prognostic Biomarker for Nasopharyngeal Carcinoma

no image available

X MING*, H Ying , R Huang , J Wang , W Hu , Z Zhang , R Zhai , f kong , X Ou , C Hu , X He , C Shen , X Wang , C Du , t Xu , Fudan University Shanghai Cancer Center, Shanghai, shanghai


TU-FG-605-4 (Tuesday, August 1, 2017) 1:45 PM - 3:45 PM Room: 605

Purpose: The purpose of this study aimed to develop prognosis signatures through radiomics analysis for patients suffering nasopharyngeal carcinoma (NPC) based on their first diagnosis magnetic resonance imaging (MRI).

Methods: 208 radiomics features of each 303 patients with stage I-IV were involved in the study. Patients were split into training cohort of 200 patients and validation cohort of 103 patients according to their first diagnosis date order. Radiomics feature analysis consists of non-supervised cluster analysis according to patient feature patterns and prognosis model building for disease free survival (DFS), overall survival (OS), distant metastasis free survival (DMFS) and local recurrence free survival (LRFS) by dimension reduction and multivariable regression on the training cohort. Also another two prognosis models based on clinical features only and combination of radiomics and clinical features were generated by the same way to estimate the incremental prognostic value of radiomic features by C-index for each event.

Results: Patients were clustered by Non-negative matrix factorization (NMF) into two groups which showed high correspondence with patients T stage (p<0.00001) and overall stage information (p<0.00001) by chi-square test. Also there was a significant difference in DFS (p=0.0052), OS (p=0.033), and LR (p=0.037) in two clustered groups except DM (p=0.11) by Log-Rank test. Radiomics signatures, incorporating both radiomics and clinical features, estimated DFS with a C-index of 0.753 [0.642, 0.865] and 0.773 [0.621, 0.924] for OS respectively on the validation cohort, and improved the prediction accuracy with 0.035 for DFS and 0.035 for OS compared with clinical features only. Nomogram for either radiomics signature was also constructed.

Conclusion: Radiomics signature for either DFS or OS developed in our study gave excellent prognostic estimation for NPC patients with a simple and noninvasive way of MRI, and then can provide more information to precise treatment decision.

Contact Email: