Aug 2026· Biomedizinische Technik. Biomedical engineering· 0 citations· 34 references
Medicine
TL;DR
Experimental results show that TDU-Net outperforms other methods in both segmentation accuracy and generalization, and significantly improving clinical diagnosis efficiency and accuracy.
Abstract
Abstract Objectives To address the challenges of complex feature variations and unclear boundary definitions between segmented targets and surrounding regions in medical images, a novel segmentation model based on Deformable Large Kernel Convolutional Attention (D-LKA) and Transformer is proposed. Methods The model first uses Vision Transformer as the encoder to enhance the ability to capture global information, overcoming the limitations of convolutional neural networks’ receptive field. In the decoder, a D-LKA decoder with deformable large kernel convolution attention is used, allowing the model to adapt to complex target features. Finally, the TRR module is introduced to coordinate information transfer between the convolutional neural network and Transformer, reducing semantic loss. Results The model is trained, validated, and tested on the Kvasir-SEG colon polyp dataset, with multiple ablation experiments. To validate generalization, experiments are also conducted on the CVC-ClinicDB dataset. Experimental results show that TDU-Net outperforms other methods in both segmentation accuracy and generalization. Conclusions TDU-Net achieves excellent segmentation results, addressing small target and incomplete feature extraction issues, significantly improving clinical diagnosis efficiency and accuracy.
Brain tumor segmentation from MRI is clinically critical yet challenging due to heterogeneous appearance and irregular boundaries. Conventional CNN based methods lack effective global context modeling, while transformer-based approaches are computationally expensive and unstable on limited datasets. To address these, we propose MMA-UNet, a novel hybrid architecture that integrates multiple complementary mechanisms within an encoder-decoder framework. The model employs an EfficientNet-B5 encoder, MedNeXt bridge blocks, a Mamba-based VSS bottleneck, a CBAM enhanced decoder, and deformable refinement for precise boundary adaptation. The model achieves a Dice score of 0.9063, IoU of 0.8304, precision of 0.9043, recall of 0.9092, and specificity of 0.9982 on FigShare benchmark, outperforming the evaluated U-Net, Attention U-Net, TransUNet and Swin UNet under the adopted experimental protocol and maintaining parameter efficiency (30.41 M), demonstrating strong robustness on T1-weighted contrast-enhanced MRI. The proposed model operates on independent 2D slices without volumetric context, and its generalizability to larger datasets remains to be established.
Breast cancer ultrasound images present challenges such as heterogeneous lesion morphologies, indistinct boundaries, and class imbalance. To overcome these difficulties, we propose an improved U-Net-based deep learning model for image segmentation. Specifically, we employ ResNet-34 as the encoder to form a ResNet34-UNet, which enhances multi-scale feature extraction and facilitates gradient flow through residual connections. In this architecture, the initial 7×7 convolution is replaced by three cascaded 3×3 convolutions. This preserves the original receptive field size but achieves gains in parameter efficiency and nonlinear modeling capacity. Furthermore, a hybrid loss function combining Weighted Binary Cross-Entropy and Dice loss is introduced to improve the detection of small lesions and the precision of boundary segmentation. The proposed method is evaluated on two public breast ultrasound datasets. Experimental results show that it significantly outperforms the standard U-Net and the original ResNet baseline in terms of accuracy, precision, recall, and Intersection over Union (IoU). The proposed approach provides an effective solution for automated, high-precision segmentation of breast ultrasound images, demonstrating considerable potential for clinical translation.
To address the challenges of blurry lesion boundaries, unstable training with small samples, and insufficient cross-domain generalization in brain tumor MRI image segmentation, this paper proposes BiomedCLIP-Seg, a novel segmentation method based on the biomedical vision-language pre-training model BiomedCLIP. Taking T2-FLAIR axial slices as input, the method introduces medical text semantic priors into the pixel-level lesion localization process through structured medical prompts and zero-initialized soft prompt mechanisms. Meanwhile, a cross-layer visual feature fusion module is designed to integrate shallow boundary textures, mid-level regional contexts, and high-level lesion semantics. Furthermore, a QKV cross-modal segmentation decoder is constructed to achieve explicit interaction between visual patch representations and category text semantics. To alleviate the issues of peripheral missed segmentation and region adhesion caused by weak tumor boundaries, a cross-modal consistency loss and an L2 norm-based boundary constraint loss are further introduced. Experimental results on the BraTS 2021 dataset demonstrate that BiomedCLIP-Seg achieves Dice, IoU, and HD95 scores of 0.903±0.004, 0.823±0.005, and 5.42±0.27 mm, respectively, outperforming representative methods including U-Net, TransUNet, CLIPSeg, nnU-Net, Swin UNETR, MedSAM, and SAMed. Ablation studies and hyperparameter sensitivity analyses further validate the effectiveness of structured medical prompts, soft prompt adaptation, cross-layer feature fusion, cross-modal consistency constraints, and boundary losses in improving model performance. This study provides a new technical pathway for utilizing biomedical vision-language models to tackle weak boundary segmentation problems in medical imaging.
Mengyuan Cao, Jialu Zhao, Xinxin Song· Intelligent & Human Futu...· 0 citations
Accurate localization and segmentation of colorectal polyps are of critical importance for the diagnosis and treatment of colorectal cancer. Given the high demand for precise polyp segmentation in this context, the Segment Anything Model (SAM) has recently demonstrated remarkable potential in the field of medical imaging. However, due to the domain differences between natural and medical images, it faces segmentation challenges arising from the variability in polyp size and shape, coupled with boundary ambiguity. To address these challenges, we propose a SAM-based boundary-aware multi-channel fusion network named SAM-BoMu, which consists of a Multi-Channel Encoder (MCE) and a Multi-Fusion Decoder (MFD). The MCE consists of a ViT encoder, a CNN encoder, and a boundary extractor to extract low-frequency, high-frequency, and boundary features. To reduce the training cost of the ViT encoder, we design an Optimization-Aid Framework for Transformer (OAFT), which only introduces trainable modules to learn spatial and frequency domain details while freezing other parameters. Subsequently, we introduce a Boundary-Constrained Hybrid Fusion (BCHF) module into the MFD to utilize these features for enhancing the segmentation performance. We systematically evaluated the performance of SAM-BoMu on five public datasets. Experimental results show that SAM-BoMu achieves competitive SOTA-level performance across all benchmarks. It obtains the best Dice of 0.954 on CVC-ClinicDB. On a private dataset (AHJU-DB) provided by the Affiliated Hospital of Jiangnan University, it still attained a Dice coefficient of 0.943, further demonstrating the outstanding generalization capability of our approach.
Shengming Gu, Xiaomao Fan, Huiming Tu et al.· IEEE journal of biomedical a...· 0 citations
Results indicate the WVM-UNet architecture effectively captures discriminative features for precise medical image segmentation, and demonstrates the competitive performance of the method on multiple public datasets.
Yulong Yang, Wenchao Gao, Zheng-Guo Wu 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
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
