UnDA framework enables cross-modal knowledge transfer in medical imaging without paired data

Researchers have introduced UnDA, a novel framework designed for unpaired cross-modal knowledge distillation in medical imaging. This approach aims to improve downstream tasks by transferring knowledge between different modalities, even when paired data is unavailable. UnDA incorporates an alignment module that extracts structured class tokens and utilizes Uncertainty-Weighted Optimal Transport (UCT-OT) to dynamically weight feature-level alignment based on prediction confidence, thereby mitigating noise from uncertain source predictions. Additionally, a per-class ProtoNCE objective is employed to maintain stable prototype memories for global discriminability across unpaired batches. Evaluations on segmentation tasks under strictly unpaired conditions have demonstrated consistent improvements in accuracy and boundary precision in the target modality. AI

IMPACT Enables more robust medical image analysis by facilitating knowledge transfer across different imaging modalities without requiring paired data.

RANK_REASON The cluster contains a research paper detailing a new method for medical imaging analysis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• Litmaps
• ProtoNCE
• ScienceCast
• scite Smart Citations
• Shadab Tanjeed Ahmad
• Uncertainty-Weighted Optimal Transport
• UnDA

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →