BBR-Net: Boundary-Balanced Replay for Continual Medical Image Segmentation
Researchers have developed a new method called Boundary-Balanced Replay Network (BBR-Net) to address the challenge of continual learning in medical image segmentation. This approach focuses on preserving anatomical structures by prioritizing replay samples that are rich in boundary information and class balance. Experiments on CAMUS and CardiacNet datasets demonstrated that BBR-Net effectively maintains performance on source tasks while reducing catastrophic forgetting and improving adaptation to new tasks. The study also highlighted that the effectiveness of replay methods is significantly influenced by the structural reliability of the stored data, rather than just memory capacity. AI
IMPACT Improves knowledge retention in AI models for medical image segmentation, potentially leading to more robust diagnostic tools.