Back to all papers

Interpretable multimodal MRI radiomics for predicting neoadjuvant chemotherapy response in nasopharyngeal carcinoma.

November 25, 2025pubmed logopapers

Authors

Lin TC,Wang KJ,Li WZ,Wu G,Chen F,Huang WY

Affiliations (4)

  • Department of Radiology, Hainan Affliated Hospital of Hainan Medical University (Hainan General Hospital), NO. 19, Xiuhua St, Xiuying Dic, Haikou, Hainan, 570311, P.R. China.
  • Department of Radiotherapy, Hainan Affliated Hospital of Hainan Medical University (Hainan General Hospital), Haikou, Hainan, P.R. China.
  • Department of Radiology, Hainan Affliated Hospital of Hainan Medical University (Hainan General Hospital), NO. 19, Xiuhua St, Xiuying Dic, Haikou, Hainan, 570311, P.R. China. [email protected].
  • Department of Radiology, Hainan Affliated Hospital of Hainan Medical University (Hainan General Hospital), NO. 19, Xiuhua St, Xiuying Dic, Haikou, Hainan, 570311, P.R. China. [email protected].

Abstract

To investigate the value of multimodal MRI radiomics model in predicting Neoadjuvant chemotherapy (NAC) response in Nasopharyngeal carcinoma (NPC) and assess the performance differences between machine learning and deep learning models. 370 NPC patients were retrospectively enrolled, with 126 patients receiving additional Dynamic contrast-enhanced Magnetic Resonance Imaging (DCE-MRI). Radiomics features were extracted from T1WI, PDWI, CE-T1WI and DCE-MRI sequences. Feature selection was performed through Mutual Information Criterion (MIC), Least Absolute Shrinkage and Selection Operator (LASSO) algorithms, and feature selection network. Machine learning models based on logistic regression, support vector machines, random forests, and extreme gradient boosting, as well as deep learning prediction models of multi-layer perceptrons were established. Model performance was evaluated using Receiver Operating Characteristic (ROC) curves and interpreted through SHapley Additive exPlanations (SHAP) values. The deep learning model incorporating DCE-MRI features showed superior performance with an Area under the curve(AUC) of 0.885, sensitivity of 79.6%, and specificity of 83.7%. SHAP analysis revealed that DCE-MRI features, particularly from late enhancement phases, contributed most significantly to model predictions. The multimodal MRI radiomics model, especially when combined with DCE-MRI, can effectively predict NAC response in NPC patients. The model's interpretability through SHAP analysis demonstrates its potential for clinical application.

Topics

Journal Article

Ready to Sharpen Your Edge?

Join hundreds of your peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.