Objective To investigate the prediction value of multi-modal MRI radiomics model in the efficacy of chemoradiotherapy for nasopharyngeal carcinoma.

Methods The clinical and MR imaging data of 151 patients with nasopharyngeal carcinoma confirmed by pathology before and after chemoradiotherapy were collected. The patients were divided into the effective treatment group(107 cases) and ineffective treatment group(44 cases) according to the solid tumor efficacy evaluation criteria(RECIST), and randomly divided into the training set and the test set according to the ratio of 7∶3. First, three sequences of lipid-pressure T2WI, diffusion-weighted imaging (DWI) and enhanced T1WI (CE-T1WI) before treatment were mapped successively as ROI, and then the image omics features were extracted. After the minimum-maximum normalization preprocessing and optimal feature screening(number) dimensionality reduction, the best image omics features were selected to construct the logistic regression model of CE-T1WI, T2WI, DWI and three sequential images. The ROC curve was used to evaluate the effects of each model predicting the efficacy of nasopharyngeal cancer. The predictive performance and benefit of the model were evaluated by calibration curve and decision curve, respectively.

Results The predictive performance of joint model was the best, the AUC of the training and test groups were 0.835 and 0.823, respectively. The specificity, sensitivity, and accuracy in the training group were 80.6%, 75.7%, and 79.0%, respectively, while the specificity, sensitivity, and accuracy in the test group were 84.8%, 72.7%, and 78.3%, respectively. The joint model had the highest predictive performance and good fit. All models of decision curve could achieve good clinical benefits.

Conclusions Multi-modal MRI radiomics model can effectively predict the efficacy of chemoradiotherapy for nasopharyngeal carcinoma, and the effect of joint model is the best.