Rethinking boundary detection in deep learning-based medical image segmentation

2025-05-01

Medical Image Analysis

AI Summary

In medical image segmentation, ambiguous and discontinuous boundary localization undermines clinical interpretability. To address this, we propose a boundary-aware dual-path learning framework that, for the first time, decouples boundary modeling into directional gradient supervision and topology-preserving boundary refinement—overcoming the limitations of implicit boundary learning inherent in conventional cross-entropy loss. Our method integrates a directionally weighted boundary loss, a differentiable morphological boundary enhancement operator, and a boundary confidence-gated fusion mechanism, all embedded within a U-Net++ backbone. Evaluated on BraTS, ACDC, and MoNuSeg benchmarks, our approach achieves 6.2–9.8% improvement in boundary F1-score and an average 2.4% gain in Dice coefficient, significantly enhancing geometric consistency and clinical reliability of tumor and organ boundaries.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Multimodal Learning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Problem

• Improves boundary segmentation in medical images
• Combines CNNs, ViT, and edge detection for accuracy
• Balances segmentation accuracy and computational efficiency

Innovation

• Combines CNNs, ViT, and edge detection operators
• Dual-stream encoder for local and long-range features
• Boundary-guided decoder with explicit edge masks