MRI-Based Radiomics Prediction Of Adverse Prognosis For Nasopharyngeal Carcinoma After Radiotherapy

Local recurrence and radiation induced temporal lobe injury are two of the major adverse outcomes after radiotherapy for nasopharyngeal carcinoma patients.This study tried to predict these two adverse outcomes using radiomics methods in patients with advanced nasopharyngeal carcinoma after radiotherapy.We aimed to assist in developing individualized treatment plan and reduce adverse events after radiotherapy in patients with advanced nasopharyngeal carcinoma.This study is mainly divided into the following two parts:1.To initially predict the risk of local recurrence in patients with locally advanced nasopharyngeal carcinoma within 5 years after radiotherapy using radiomics method;2.Using radiomics method to predict the risk of radiation temporal lobe injury in patients with T4 nasopharyngeal carcinoma within 5 years after radiotherapy.Part Ⅰ:MRI-base Radiomics for The Prediction of Local Recurrence of Advanced Nasopharyngeal CarcinomaObjective To develop and validate a prediction model incorporating radiomic features with clinical variables for individual local recurrence risk assessment in advanced nasopharyngeal carcinoma（NPC）patients before initial treatment.Methods MRI and clinical data of nasopharyngeal carcinoma patients who received intensity modulated radiotherapy in our hospital from January 2010 to December 2019 were retrospectively collected.Three sequences of T1-weighted（T1WI）,T2-weighted（T2WI）and T1-contrast enhanced（CET1）images were extracted before treatment.The minimum absolute contraction and selection operator（LASSO）regression method was used to select the radiomics features.Support vector machine,random forest,decision tree and other machine learning methods were used to establish prediction models to predict the risk of local recurrence of advanced nasopharyngeal carcinoma within 5 years after radiotherapy.Receiver operating characteristic curve（ROC）and Area Under Curve（AUC）were used to evaluate the prediction efficiency.Results A total of 809 nasopharyngeal carcinoma patients with regular follow-up were retrospectively collected,and 246 patients with locally advanced nasopharyngeal carcinoma were included（85 patients with local recurrence and161 patients without local recurrence）.A total of 2862 radiomics features（954features per sequence）were preliminarily extracted from T1-weighted image（T1WI）,T2-weighted image（T2WI）and T1-contract enhanced（CET1）image.After selected by LASSO method,58 image features were reserved for model prediction.The AUC of T1WI,T2WI and CET1 sequences was 0.8386,0.8053and 0.7535,respectively.The combination of T1WI+T2WI+CET1 sequences showed the best predictive result with an AUC of 0.8817 and F1score of 0.8654.The selection of the clinical factors showed that family history of nasopharyngeal carcinoma,age,sex,absolute value of lymphocytes,proportion of lymphocytes and distribution width of platelet volume had influence on local recurrence of nasopharyngeal carcinoma after radiotherapy.The prediction of local recurrence in patients with advanced nasopharyngeal carcinoma within 5years after radiotherapy using imaging features combined with clinical factors showed better results.The predictive AUC of T1WI+clinical factors,T2WI+clinical factors,CET1+clinical factors,T1WI+T2WI+clinical factors,T1WI+CET1+clinical factors,and T1WI+T2WI+CET1+clinical factors were83.8%,83.2%,80.8%,87.7%,84.3%,85.5%,91.1%respectively in the training set,and 79.2%,77.1%,85.4%,76.1%,85.0%,82.0%,75.3%respectively in validation set.Conclusions In this study,a prediction model of local recurrence risk of advanced nasopharyngeal carcinoma within 5 years after radiotherapy was established based on the pre-treatment MR imaging and clinical features.MRI features based on multiple sequences can predict local recurrence in patients with advanced nasopharyngeal carcinoma within 5 years after radiotherapy.Part Ⅱ:Prediction of Radiation-induced Temporal Lobe Injury in T4 Stage Nasopharyngeal Carcinoma with MRI-base RadiomicsObjective This study aims to use pretreatment MRI-based radiomics data with clinical data to predict radiation-induced temporal lobe injury（RTLI）in nasopharyngeal carcinoma（NPC）patients with stage T4/N0–3/M0 within 5years after radiotherapy.Methods This study retrospectively examined 98 patients（198 temporal lobes）with stage T4/N0–3/M0 NPC.Participants were enrolled into a training cohort or a validation cohort in a ratio of 7:3.Radiomics features were extracted from pretreatment MRI that were T1-w and T2-w.Spearman rank correlation,the t-test,and the least absolute shrinkage and selection operator（LASSO）algorithm were used to select significant radiomics features,and machine learning models were used to generate radiomics signatures（Rad-Scores）.Rad-Scores and clinical factors were integrated into a nomogram for prediction of RTLI.Nomogram discrimination was evaluated using receiver operating characteristic（ROC）analysis,and clinical benefits were evaluated using decision curve analysis（DCA）.Results Participants were enrolled into a training cohort（n=139）or a validation cohort（n=59）.A total of 3568 radiomics features were initially extracted from T1-w and T2-w images.Age,D_max,D_1cc,and 16 stable radiomics features（6 from T1-w and 10 from T2-w images）were identified as independent predictive factors.A greater Rad-Score was associated with greater risk of RTLI.The nomogram showed good discrimination,with a C-index of 0.85（95%CI:0.79–0.92）in the training cohort and 0.82（95%CI:0.71–0.92）in the validation cohort.Conclusion We developed models for prediction of RTLI in patients with stage T4/N0–3/M0 NPC using pretreatment radiomics data and clinical data.Nomograms from these pretreatment data improved the prediction of RTLI.These results may allow the selection of patients for earlier clinical interventions.