Sep 2026· IEEE Internet of Things Journal· Vol 13, pp. 41034-41042· 0 citations· 37 references
Abstract
Accurate medical image segmentation remains challenging due to complex anatomical structures and ambiguous boundaries. To address the loss of structural cues in existing methods, we propose ContourFD-Net, a gradient-guided attention network. Unlike existing contour-aware networks that rely on auxiliary supervision, ContourFD-Net introduces an explicit structural prior via fixed finite-difference operators and couples gradient modeling with attention-based feature refinement. This approach preserves fine details through a dual-path architecture and decoupled spatial–channel attention. Extensive experiments on four benchmark datasets demonstrate the effectiveness and robustness of the proposed method. On the DSB 2018 dataset, ContourFD-Net achieves a Dice coefficient of 91.67%, demonstrating superior performance compared with representative state-of-the-art methods. Moreover, the framework requires only 6.40 M parameters and 3.39 GFLOPs, demonstrating a favorable accuracy–efficiency tradeoff. On the NVIDIA Jetson Orin NX platform, ContourFD-Net achieves an average inference latency of 16.97 ms and a throughput of 59.05 queries per second (QPS) with only 49-MiB memory consumption, validating its efficiency for real-time deployment on resource-constrained edge devices. The source code and model weights are available at https://github.com/ZBKim/ContourFD-Net
MRD-UNet provides a practical balance between segmentation accuracy and computational efficiency and outperforms baseline CNNs and performs comparably to heavier transformer-based models while using significantly fewer parameters.
Musa Doğan, I. Ozkan· BMC Medical Imaging· 0 citations
Few-shot Medical Image Segmentation (FS-MIS) has garnered increasing attention for its ability to reduce reliance on large-scale pixel-wise annotations. How ever, most existing methods rely solely on single-level encoder outputs, neglecting the complementary roles of hi erarchical features in contour refinement and regional dis crimination. This limitation often results in boundary ambi guity and local segmentation errors. To this end, we pro pose a Triple-Attention Coordinated Cross-layer Dynamic Network (TriCD-Net) for FSMIS. Specifically, we design a Feature-Enhanced Boundary Refinement (FEBR) module in each encoder layer, where three parallel attention branches explicitly model foreground, background, and boundary regions. The resulting features are passed to a Dense Dual-Contrast Segmentation (DDCS) head to perform segmenta tion, enabling progressive boundary refinement. Consider ing that query images inherently contain valuable structural information, we introduce a mask reconstruction task and design a Cross-layer Dynamic Convolution Fusion (CDCF) module that dynamically generates convolution kernels to adaptively fuse multi-level features for query reconstruc tion. Furthermore, we establish an interaction between the reconstruction and segmentation branches through an Uncertainty-Gated Cross-Attention Injection (UCI) mod ule, which selectively injects reconstruction features into the segmentation branch to enhance structural coher ence and boundary accuracy. Extensive experiments on three public benchmark datasets demonstrate that TriCD-Net consistently achieves state-of-the-art performance. The code is available at https://github.com/qchi-code/TriCD-Net.
Qiang Chi, Fuzhi Wu, Yuhang Liu et al.· IEEE Transactions on Medical...· 0 citations
Early-pregnancy ultrasound images often contain gestational sac, yolk sac, and embryo regions with blurred boundaries, variable shapes, and small target sizes, which makes accurate segmentation difficult. In addition, many existing segmentation networks have high computational cost and are not well suited for resource-constrained clinical devices. To address these challenges, we propose EFAU-Net, an efficient feature-enhanced U-shaped network for segmenting key structures in early-pregnancy ultrasound images. EFAU-Net introduces three components: a grouped channel shuffle fusion (GCSF) module to improve feature utilization in deep layers with low computational overhead, a local attention enhancement (LAE) module to strengthen multi-scale feature fusion and small-target representation, and a ground-truth boundary generation (GBG) module to enhance boundary supervision and improve contour alignment. Experiments show that EFAU-Net achieves Dice scores of 97.14%, 91.25%, and 81.83% for gestational sac, yolk sac, and embryo segmentation, respectively, with 0.826 M parameters for each target-specific binary segmentation model. Compared with representative baseline and competing methods, the proposed model provides a better balance between segmentation accuracy and model complexity. These results indicate that EFAU-Net is effective for segmenting key anatomical structures in early-pregnancy ultrasound images. In the supplementary video-level biometric measurement analysis, EFAU-Net was further applied frame by frame to retained ultrasound videos, and the maximum measured value across frames was used as the video-level result, providing a useful basis for subsequent quantitative image analysis.
Danling Cheng, Jiakai Wang, Tingting Dong et al.· Journal of King Saud Univers...· 0 citations
This paper proposes a novel framework that effectively leverages unlabeled data to improve segmentation performance in cardiac structures and applies a novel consistency constraint by a dual fine-grained boundary loss that provide global characteristics-based guidance from the transition of the boundary region and an edge-aware uncertainty loss.
Waqas Anwaar, Van Manh, Wufeng Xue et al.· Interdisciplinary Sciences C...· 0 citations
Segmenting brain tumors from MRI scans is a challenging aspect of medical image analysis because of anatomical complexity, ambiguous tumor boundaries, and variability of shape. In this paper, we propose W-AGRU-Net, a new dual-stream U-Net framework that combines W-Attention mechanisms with residual connections for strong and accurate glioma segmentation. Our framework uses two asymmetric streams that have pyramid-shaped dilated convolution schemes (PDCS) for hierarchical multi-scale context capturing. The framework also includes residual units that utilize Squeeze-and-Excitation (SE) blocks for channel-wise recalibration of features across multiple resolutions, thereby strengthening relevant tumor regions while reducing noise. The method is thoroughly evaluated on the TCIA LGG Segmentation and Figshare datasets, where it outperformed the state-of-the-art approaches. On the Figshare dataset, the approach achieved best-in-class metrics, including a registration Dice coefficient of 97.87% and Jaccard index of 97.44%, as well as 95.14% precision and 93.42% sensitivity. On the TCIA dataset, we achieved 94.0% Dice, 99.9% pixel-level accuracy, 89.29% Jaccard index, and 90.01% sensitivity, demonstrating improved performance over baseline methods. Concerning more recent state-of-the-art approaches with Dice scores of 92.0% (TCIA) and 96.9% (Figshare), our approach showed enhanced accuracy in brain tumor segmentation, and we expect our approach to work well for clinical brain imaging applications.
Marwa Abbas, Hussein Mogahed, A. Hussein et al.· Scientific Reports· 0 citations
Medical ultrasound (US) image segmentation faces significant challenges due to speckle noise, low-contrast boundaries, acoustic shadowing, and acquisition variation across operators and clinical centers. Although encoder-decoder and transformer-based networks have achieved strong performance, many methods recover boundary details through dense decoders or larger backbones, which may still produce over-smoothed contours or unstable predictions under external distribution shifts. In this article, we propose Risk-routed Implicit Boundary Refinement (RIBR), a compact segmentation framework that uses implicit neural representation as a risk-routed residual correction rather than an unconstrained full-mask predictor. RIBR combines boundary-refinement implicit residuals, risk-routed residual control, and geometry- and speckle-aware boundary regularization to refine uncertain contours while suppressing non-boundary oscillations. Evaluation on nine US datasets covering lymph nodes, breast lesions, thyroid nodules, and prostate shows that RIBR achieves the best overall macro-average and consistently reduces boundary error across grouped and organ-specific comparisons under a compact parameter budget. These findings suggest that controlled implicit residual learning is a practical strategy for resource-constrained and boundary-sensitive US segmentation. Source code is available at https://github.com/jinggqu/ribr.
Jingguo Qu, Xinyang Han, Xiang Wang et al.· 0 citations