Jul 2026· Journal of Computational Design and Engineering· Vol 13, pp. 196-211· 0 citations
TL;DR
SegRWKV, a hybrid architecture that integrates pretrained Vision-RWKV modules in the encoder for efficient global dependency modeling, and a PixelRefinement module in the decoder to improve feature reconstruction and multi-scale alignment, is proposed, highlighting SegRWKV’s superior ability to balance accuracy, efficiency, and scalability.
Abstract
Medical image segmentation is critical for clinical diagnosis, treatment planning, and prognosis. However, practical deployment of segmentation models is often hindered by two main challenges: (1) inefficient global dependency modeling, which slows down context integration and limits feature expressiveness; and (2) insufficient decoder feature reconstruction and multi-scale alignment, which prevents accurate recovery of fine anatomical details. These challenges are exacerbated in high-resolution medical imaging, where computational complexity and slow inference speed further limit model effectiveness. Current solutions, including CNN-based, Transformer-based, and state-space-based models, fail to simultaneously balance efficiency, fine-detail reconstruction, and scalability.To address these limitations, we propose SegRWKV, a hybrid architecture that integrates pretrained Vision-RWKV (VRWKV) modules in the encoder for efficient global dependency modeling, and a PixelRefinement module in the decoder to improve feature reconstruction and multi-scale alignment. By using recursive spatial mixing and channel-wise decoupled projections, SegRWKV reduces computational overhead while maintaining high feature expressiveness. A lightweight variant, SegRWKV-S, reduces model size by decreasing channel width and convolutional depth, achieving compression without sacrificing accuracy.Comprehensive experiments on several medical imaging datasets, including AbdomenMRI, Endoscopy, and Microscopy, demonstrate that SegRWKV consistently outperforms CNN-, Vision Transformer-, Mamba-, and RWKV-UNet-based baselines. Specifically, SegRWKV-B improves DSC/F1 scores by 4–22% over CNNs, 8–58% over Transformers, 1–8% over Mamba, and 10–27% over RWKV-UNet. Efficiency analysis shows that SegRWKV delivers 4.8 × faster inference and reduces GPU memory usage by 4.5 × compared to existing architectures, with stable computational scaling as input resolution increases. These results highlight SegRWKV’s superior ability to balance accuracy, efficiency, and scalability, particularly in high-resolution medical image segmentation tasks, where many attention-based models struggle to scale effectively. Code and pretrained models are available at: https://github.com/luminescentfrr/SegRWKV/.
Segmentation of medical images is a crucial process for diagnosis and treatment planning. Yet, traditional CNN based models are often not able to convey high-level spatial relationships and intricate tissue boundaries in high-resolution medical images. To tackle these issues, this paper presents a novel ResUNet++–Vision Transformer (ResUNet++–ViT) framework incorporating both multi-scale local feature extraction and global contextual learning. The ResUNet++ backbone consists of residual blocks and nested skip connections, which are used to extract hierarchical features, and the Vision Transformer uses self-attention mechanisms to capture long-range dependencies. A fusion module allows for the integration of local and global features, resulting in better segmentation accuracy and preserving the boundary. The proposed model was tested on the ISBI 2012 Electron Microscopy Segmentation Challenge (EMSC) dataset, and obtained a Dice score of 0.960, IoU of 0.920, precision of 0.967, recall of 0.958, accuracy of 0.984 and Hausdorff distance of 2.76. The proposed framework is compared with FCN, UNet, Attention UNet, ResUNet, UNet++, Vision Transformer and TransUNet, and the results show its superiority. The results show that the combination of ResUNet++ and Vision Transformers greatly enhances the performance of segmentation, boundary delineation, and generalization in the field of advanced medical image analysis applications.
G. Satyanarayana, Kadali Satyanarayana, Renuka Kondabala et al.· 2026 International Conferenc...· 0 citations
TvaraNet is pro-posed, an extremely lightweight segmentation network designed to preserve boundary fidelity under strict efficiency constraints and achieves competitive or superior boundary-aware performance compared to heavier architectures.
Reliable medical image segmentation remains challenging because models must preserve fine boundary details while maintaining global semantic consistency. CNNs capture local structures effectively but have limited long-range modeling ability, whereas Transformer-based methods improve global context at high computational cost. Mamba-based state space models offer efficient long-range modeling, but may weaken high-frequency textures and boundary cues. To address these limitations, we propose DIG-MambaNet, a Dual-path Interactive Guided Mamba Network for medical image segmentation. The network introduces a dual-path complementary modeling block (DCM Block), where a cross-feature spatial interaction module (CSIM) adaptively integrates CNN-based local features and Mamba-based global features. A source image-guided module (SIGM) injects high-frequency information from the original image to compensate for downsampling-induced detail loss, while an inter-layer detail refinement fusion module (IDRFM) improves encoder–decoder feature alignment during reconstruction. Experiments on 2018DSB, ISIC2018, JSUAH-Cerebellum, and CVC-ClinicDB, covering nuclei segmentation in microscopy images, skin lesion segmentation in dermoscopic images, fetal cerebellum segmentation in ultrasound images, and polyp segmentation in colonoscopy images, demonstrate that DIG-MambaNet achieves consistent and competitive performance across diverse target structures and imaging conditions, with improved boundary delineation and favorable overlap-based accuracy compared with representative CNN-, Transformer-, and Mamba-based methods.
Yongkang Zhu, Tianyu Yu, Hongmei Li et al.· Journal of Imaging· 0 citations
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
Accurate diagnosis in medical imaging is often hampered by two intrinsic factors: the complex, anomalous geometric distortion of anatomical structures and the extreme size variability of pathological lesions. Existing convolutional neural networks (CNNs) struggle with global context, while Vision Transformer (ViT) models are limited by quadratic computational costs or reliance on window-based attention mechanisms that disrupt semantic continuity. To address these limitations, we propose MaxStripViT, a novel hybrid architecture that effectively integrates distortion modeling with multi-axis strip attention mechanisms and Local Position-Aware blocks. Our method introduces three key contributions: 1) geometric-adaptive stem (GAS) leverages learnable offsets via Deformable Convolutions (DCNv2) to dynamically align the sampling grid with irregular organ boundaries at the earliest feature extraction stage. It effectively mitigates background noise; 2) position-aware local block is proposed to enhance the Mobile Inverted Bottleneck (MBConv) with Coordinate Attention. This mechanism explicitly models long-range dependencies along spatial axes, improving the precise localization of subtle lesions; and 3) novel sequential multi-axis strip attention mechanism is proposed to replace shift window-based self-attention. The proposed method is experimented on three dataset benchmarks of CT and ChestXray imaging for multi-class lung disease such as CheXtImageNet, IQ-OTH/NCCD, and ChestXray-Image dataset. The results demonstrated that MaxStripViT achieves robust performance and outperforms standard ViT models, and improves performance compared to the most advanced hybrid models currently available, including Swin Transformer, MaxViT, CSWin, ConvNeXt, ConvNeXtv2, and EfficientNet-B7, in terms of classification accuracy and computational efficiency. The proposed method provides a robust solution for health scenarios requiring both geometric flexibility and multi-scale interpretability.
Thanh-An Pham, Tuong-Lan Le van, Van-Dung Hoang· IEEE Access· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
