Multimodal deep learning predicts induction chemo response and survival in nasopharyngeal cancer

A multi-task model (MoEMIL) integrating MRI and pathology images achieved an AUC of 0.801 for predicting induction chemotherapy response in locally advanced nasopharyngeal carcinoma, significantly outperforming radiomics and TNM staging. It also stratified overall survival (P<0.…

• The poor responder group consisted exclusively of patients with stable disease; no cases of progressive disease were observed, which may limit generalizability.
• Compared with a Pathomics model, MoEMIL significantly improved overall survival prediction (P=0.0049) but showed only borderline improvement for induction chemotherapy response (P=0.064).

Automated summary

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