Multiparametric MRI radiomics and 2.5D deep learning predict early complete response to induction chemoimmunotherapy in nasopharyngeal carcinoma

In a retrospective study, a combined MRI radiomics+2.5D deep learning model predicted early complete response to chemoimmunotherapy in nasopharyngeal cancer with an external test AUC of 0.821, sensitivity 0.859, and NPV 0.932.

• Design: Retrospective analysis of 500 patients (training: 195, internal validation: 84, external test: 221) from two centers; model integrated clinical, habitat radiomics, and 2.5D deep learning features from pretreatment T1w, T2w, and contrast-enhanced MRI.
• Key secondary: SHAP analysis showed both 2.5D deep learning and habitat radiomics features contributed prominently to predictions.
• Limitation: Retrospective design and single external test center; prospective multicenter validation is needed before clinical adoption.

RadPigeon summaries are original and for information only. They are not clinical advice.