ACorD: Adaptive Correspondence Discovery for Medical Image Registration

Key points

• To address these challenges, we propose ACorD, an adaptive correspondence discovery framework designed to mitigate error propagation with an attention-guided process.1
• ACorD first predicts multiple potential correspondence candidates through Correlation-guided Cross-scale Offset Prediction (CCOP), which combines correlation maps as spatial matching priors with cross-scale features as global deformation context.1
• By integrating adaptive offset prediction with selective, attention-based aggregation of correspondence evidence, ACorD enables accurate deformation refinement.1
• Extensive experiments on four public medical registration benchmarks demonstrate that adaptive correspondence discovery within a multi-scale refinement framework improves registration robustness across diverse anatomical structures.1

Sources (1)

• [1]ACorD: Adaptive Correspondence Discovery for Medical Image Registration
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 01:31 AM

  Medical image registration is a fundamental task for aligning anatomical structures across images.

  To address these challenges, we propose ACorD, an adaptive correspondence discovery framework designed to mitigate error propagation with an attention-guided process.

Extractive summary: sentences quoted from the sources.