Multimodal MRI habitat-based deep learning fusion model predicts neutron therapy response in cervical cancer

An MRI-based habitat deep learning fusion model predicted response to californium-252 neutron brachytherapy in cervical cancer, achieving an AUC of 0.892 (95% CI 0.821–0.963) in the validation set, significantly outperforming clinical and radiomics models.

• Habitat analysis identified three functional subregions linked to tumor heterogeneity: high vascularity/high cellularity, low vascularity/high cellularity, and low vascularity/low cellularity.
• The fusion model (AUC 0.892) outperformed clinical-only (AUC 0.712), whole-tumor radiomics (AUC 0.783), and radiomics+deep learning (AUC 0.835) models (all P<0.05).
• model-predicted high-benefit patients had 3-year overall survival of 92.5% vs 71.4% in the low-benefit group (P=0.003).

Automated summary

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