Reliability-Guided Vision-Language Decoding with Text-Aware Skip Refinement for Medical Image Segmentation

Key points

• Medical image segmentation remains challenging when targets exhibit limited visual distinction, irregular structure, or substantial variation across imaging modalities.1
• Medical text can reduce this ambiguity, but indiscriminate fusion may introduce unreliable guidance.1
• Its Reliability-guided Fusion Gate (RFG) adaptively controls cross-modal updates, while the Text-guided Skip Refinement Adapter (TSRA) refines high-resolution features using text guidance.1
• RELiF is evaluated on radiographic, CT, and endoscopic segmentation tasks, where it shows competitive performance across diverse imaging characteristics and target appearances.1

Sources (1)

• [1]Reliability-Guided Vision-Language Decoding with Text-Aware Skip Refinement for Medical Image Segmentation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 12:29 PM

  We present RELiF, a vision--language framework that integrates text through reliability-guided decoding.

  Medical image segmentation remains challenging when targets exhibit limited visual distinction, irregular structure, or substantial variation across imaging modalities.

Extractive summary: sentences quoted from the sources.