Multimodal MRI-based radiomics model for predicting short-term efficacy in nasopharyngeal carcinoma.

Problem: Accurate prediction of short-term treatment response remains a critical challenge in nasopharyngeal carcinoma (NPC) management. Traditional TNM staging and clinical biomarkers offer limited precision for individualized therapy planning, creating a need for more robust, non-invasive predictive tools. Aim: This multicenter study aimed to develop and validate a multimodal MRI-based radiomics model for predicting short-term treatment response in NPC, and to compare its performance against conventional clinical biomarkers. Methods: We analyzed pre-treatment T1-weighted, T2-weighted, and contrast-enhanced T1-weighted MRI sequences from 173 patients in our primary cohort and 55 external validation cases. A total of 3,591 radiomic features were extracted per patient. After rigorous feature selection using maximum relevance minimum redundancy (mRMR) and Least Absolute Shrinkage and Selection Operator (LASSO) regression, we developed and compared eight machine learning classifiers. Model performance was evaluated through comprehensive validation, including calibration analysis and decision curve assessment. Results: The Support Vector Machine (SVM) model demonstrated superior performance, achieving an area under the curve (AUC) of 0.935 (95% CI: 0.867-1.000) on internal testing with balanced sensitivity (87.1%) and specificity (95.2%). External validation confirmed model robustness (AUC 0.880, 95% CI: 0.800-0.960). Our radiomics approach significantly outperformed all clinical biomarkers (AUC improvement: 18.7-24.3%, Conclusion: The multimodal MRI-based radiomics model represents a transformative non-invasive tool for predicting short-term treatment response in NPC, offering superior performance to conventional methods and providing valuable insights for personalized treatment strategies. Our findings support the integration of radiomics into clinical decision-making for NPC management.