Clinical value of MRI radiomics in predicting radiation-induced temporal lobe injury in locally advanced nasopharyngeal carcinoma
LU Zhiwei
LI Hongliang
LIU Suya
MA Yuanyuan
Abstract：Objective To explore the clinical value of magnetic resonance imaging(MRI)radiomics parameters in predicting the risk of radiation-induced temporal lobe injury(RTLI)in patients with locally advanced nasopharyngeal carcinoma(LA-NPC).Methods A retrospective analysis was conducted on 126 patients with LA-NPC who received intensity-modulated radiotherapy(IMRT)and concurrent platinum-based chemotherapy from June 2021 to June 2023.Patients were randomly assigned to the training set(n=88)and the test set(n=38)in a ratio of 7:3.Univariate and multivariate Logistic regression models were used to predict the risk factors for RTLI occurrence.The radiomics feature data in the MRI imaging sequences before treatment were extracted,and the Mann-Whitney,Pearson correlation,minimum absolute shrinkage and selection operator(LASSO)algorithms were used to screen the significant features.The clinical features and radiomics features screened out were integrated into a joint model based on the logistic regression algorithm.The receiver operating characteristic(ROC)curve,calibration curve and decision curve were used to compare and verify the predictive performance of the model.Results A total of 49 patients with LA-NPC developed RTLI,including 34 cases in the training set and 15 cases in the test set.The median latency from the completion of IMRT to the first MRI detection of RTLI in all damaged temporal lobes was 19.12 months.Logistic regression analysis showed that T stage(OR=8.414,95％CI:1.945-36.402,P=0.004)and D0.5cc(OR=5.315,1.610-17.543,P=0.006)were independent risk factors for predicting RTLI in patients with LA-NPC.The Rad-score model was constructed using six MRI texture features significantly associated with RTLI,and the combined model was constructed by combining two independent clinical factors.ROC curve analysis showed that the predictive performance of the combined model in the training set was superior to that of either the clinical model alone or the radiomics model(all P＜0.05).In the test set,the combined model superior to that of the clinical feature model(P=0.017),but showed no statistically significant difference when compared with the MRI-based radiomics model(P＞0.05).The calibration curve indicated good agreement between the predictions of the combined model and actual observations,while the decision curve demonstrated favorable clinical net benefit of the model.Conclusion The combined model constructed by integrating MRI radiomics features and clinical features can effectively predict the risk of RTLI in patients with LA-NPC and is expected to become an important auxiliary tool for RTLI prediction.
Keywords：locoregionally advanced nasopharyngeal carcinomaradiation-induced temporal lobe injurymagnetic resonance imagingradiomics featuresclinical features
Publication Date：2026-02-28
Online Publishing Date：2026-04-01(First online date of this platform, not the publication date of the document)
Pages：8( 167-174 )