Technology Category
Computer Vision: SegmentationMachine Learning: Calibration & Uncertainty QuantificationSearch and Optimization: Learning to Search
Dual-Attention U-Net++ with Class-Specific Ensembles and Bayesian Hyperparameter Optimization for Precise Wound and Scale Marker Segmentation
🤖 AI Summary
Precise segmentation of wound regions and ruler markings in clinical wound images is challenged by severe class imbalance and high morphological variability. To address these issues, we propose a dual-attention-enhanced U-Net++ architecture, employing EfficientNet-B7 as the backbone and integrating SCSE (Squeeze-and-Excitation + Channel-Spatial) modules for joint channel-spatial attention. The framework further incorporates class-specific model ensembling, Bayesian hyperparameter optimization (via Weights & Biases), and test-time augmentation to improve robustness for small objects and boundary delineation. Evaluated on the NBC 2025 and PCBBE 2025 competition benchmarks, our method achieves a weighted F1 score of 0.8640—significantly surpassing baseline approaches. This demonstrates state-of-the-art performance and strong generalization capability for complex medical image segmentation tasks.
Technology Category
Application Category
📝 Abstract
Accurate segmentation of wounds and scale markers in clinical images remainsa significant challenge, crucial for effective wound management and automatedassessment. In this study, we propose a novel dual-attention U-Net++ archi-tecture, integrating channel-wise (SCSE) and spatial attention mechanisms toaddress severe class imbalance and variability in medical images effectively.Initially, extensive benchmarking across diverse architectures and encoders via 5-fold cross-validation identified EfficientNet-B7 as the optimal encoder backbone.Subsequently, we independently trained two class-specific models with tailoredpreprocessing, extensive data augmentation, and Bayesian hyperparameter tun-ing (WandB sweeps). The final model ensemble utilized Test Time Augmentationto further enhance prediction reliability. Our approach was evaluated on a bench-mark dataset from the NBC 2025 & PCBBE 2025 competition. Segmentationperformance was quantified using a weighted F1-score (75% wounds, 25% scalemarkers), calculated externally by competition organizers on undisclosed hard-ware. The proposed approach achieved an F1-score of 0.8640, underscoring itseffectiveness for complex medical segmentation tasks.
Problem
Research questions and friction points this paper is trying to address.
Accurate segmentation of wounds and scale markers
Addressing class imbalance in medical images
Enhancing prediction reliability with ensemble models
Innovation
Methods, ideas, or system contributions that make the work stand out.
Dual-attention U-Net++ with SCSE and spatial mechanisms
Class-specific ensembles with Bayesian hyperparameter optimization
Test Time Augmentation for enhanced prediction reliability
🔎 Similar Papers
No similar papers found.
