A Multidimensional Model Based On MRI Radiomic Features,clinicopathological,and Blood Cell Parameters To Predict The Prognosis In Patients With Nasopharyngeal Carcinoma

Objective: By analyzing and screening magnetic resonance imaging(MRI)radiomic features,clinicopathological and blood cell parameters,a multidimensional prognostic model for progression-free survival(PFS)of nasopharyngeal carcinoma(NPC)was developed,which provided a decision base for precise diagnosis and treatment of NPC.Methods: A total of 462 patients with pathologically confirmed nasopharyngeal squamous cell carcinoma treated at Sichuan Cancer Hospital were recruited from 2015 to 2019.Patients were randomly divided into training cohorts(n=323)and validation cohorts(n=139)at a ratio of 7:3.Data collected from patients included axial CET1-w and T2-w MRI;clinicopathological parameters including gender,age,TNM stage,EBV DNA copies,smoking and drinking habits,and Ki-67;and blood cell parameters including routine blood parameters,neutrophil-to-lymphocyte ratio,lymphocyte-to-monocyte ratio,and platelet-to-lymphocyte ratio.MRI images were preprocessed using N4 bias field corrections and intensity adjustments,followed by a layer-by-layer segmentation of nasopharyngeal tumor edges using 3D Slicer 4.11(an open-source software program).Meanwhile,the features of radiomics were extracted,including first-order statistics features,3D shape-based features,textural features,wavelet-transform features,and Laplacian of Gaussian(Lo G)filtration features.After Z-score normalization of all radiomic features,the LASSO-Cox regression was used for features screening with PFS.Based on the screened features and their regression coefficients,a linear formula for calculating the rad-score for each patient was built.Rad-score cut-off values were determined by receiver operating characteristic curves(ROC),which used Kaplan-Meier for survival analysis to divide patients into high-and low-risk groups for progression.ROC curves were plotted for each blood cell parameter,and the area under the curve(AUC)values for each parameter were compared.Using a multivariate Cox model,we constructed the nomogram using eleven parameters,which included rad-score,Ki-67,EBV,and three blood parameters;we evaluated its effectiveness by using calibration curves and concordance indices(C-indexes),and we compared its benefit to TNM staging using decision curve analysis(DCA).R4.1.0,Python 3.8.5,and SPSS 20.0 software were used for statistical analysis.Results:1.Baseline situation and survival analysis.The median follow-up time was 32.88 months(range: 0.6-76.2 months).The median age was 49 years(12-82 years);133 cases were female,and 329 cases were male.In stages I,II,III,IVA,and IVB,respectively,there were 0,23,193,226,and 20 cases.The overall survival(OS)rates at 1-year,3-year,and 5-year were 98.5%,96.3%,and 90.1%,respectively,and their PFS rate at 1-year,3-year,and 5-year were: 97.2%,89.3%,and 85.4%.Fourteen patients developed distant metastases,eight recurred,and twenty-three died.2.Nine radiomic features were significantly associated with PFS.In total,2074 radiomics features were extracted from each axis CET1-w and T2-w MRI,and 9 features were finally screened by LASSO-Cox regression.The Kaplan-Meier survival analysis of the rad-score showed a significant difference in PFS between the two groups(P < 0.05).3.Three blood cell parameters were significantly correlated with PFS.The highest AUC values for monocytes,monocyte ratio,and mean corpuscular volume were 0.637,0.626,and 0.568,respectively.4.The multidimensional model predicted PFS significantly better than traditional TNM staging.The multidimensional prognostic model combining rad-score,EBV,Ki-67,gender,age,smoking habit,drinking habit,clinical-stage,monocytes,monocyte ratio,and mean corpuscular volume had the best predictive efficacy,with a significantly higher C-index than the TNM staging system in both cohorts(training cohort: 0.823 vs.0.610;validation cohort.0.812 vs.0.602).The multiparametric nomogram’s 3-year and 5-year calibration curves showed that their predicted PFS was close to the actual situation.In addition,the DCA curves showed that the model predicted well within a 50% threshold probability.Conclusion: This study validated that rad-score from MRI-based radiomic features was an independent prognostic factor for NPC.The multidimensional model,which constructed MRI-based radiomic features,clinicopathological,and blood parameters demonstrated good predictive performance and stability in predicting PFS in patients with NPC.