Scale-aware adaptive supervised network with limited medical annotations

2026-01-02

Pattern Recognition

AI Summary

This work addresses the challenges of semi-supervised medical image segmentation under conditions of scarce annotations, high inter-annotator variability, and insufficient multi-scale feature fusion, where existing methods suffer significant performance degradation on small structures and boundary regions. To this end, we propose SASNet, a dual-branch architecture that effectively integrates low-level and high-level features through three key innovations: a scale-adaptive reweighting strategy, a 3D Fourier-domain view variation augmentation mechanism, and a sign-distance-map-based consistency learning framework that jointly models spatial, temporal, and geometric consistency between segmentation and regression tasks. Extensive experiments demonstrate that our method substantially outperforms current semi-supervised approaches on the LA, Pancreas-CT, and BraTS datasets, achieving performance close to fully supervised baselines, with notable improvements in the accuracy of small lesion and boundary delineation.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Semi-Supervised Learning
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs

Problem

• semi-supervised learning
• medical image segmentation
• annotation scarcity
• inter-annotator variability
• multi-scale feature integration

Innovation

• scale-aware adaptive reweight
• view variance enhancement
• segmentation-regression consistency
• semi-supervised medical segmentation
• signed distance map