DAG‐Net: Dual‐Branch Attention‐Guided Network for Multi‐Scale Information Fusion in Lung Nodule Segmentation DOI Open Access

Bojie Zhang,

Hongqing Zhu, Ziying Wang

et al.

International Journal of Imaging Systems and Technology, Journal Year: 2024, Volume and Issue: 34(6)

Published: Nov. 1, 2024

ABSTRACT The development of deep learning has played an increasingly crucial role in assisting medical diagnoses. Lung cancer, as a major disease threatening human health, benefits significantly from the use auxiliary systems to assist segmenting pulmonary nodules. This approach effectively enhances both accuracy and speed diagnosis for physicians, thereby reducing risk patient mortality. However, nodules are characterized by irregular shapes wide range diameter variations. They often reside amidst blood vessels various tissue structures, posing significant challenges designing automated system lung nodule segmentation. To address this, we have developed three‐dimensional dual‐branch attention‐guided network (DAG‐Net) multi‐scale information fusion, aimed at types sizes. First, encoding structure is employed provide with prior knowledge about texture information, which aids better identifying different Next, designed extract global network's ability localize sizes fusing multiple resolutions. Following that, fused parallel used attention mechanisms guide suppressing influence non‐nodule regions. Finally, attention‐based achieving more accurate segmentation progressively using high‐level semantic each layer. Our proposed achieved DSC value 85.6% on LUNA16 dataset, outperforming state‐of‐the‐art methods, demonstrating effectiveness network.

Language: Английский

PMFSNet: Polarized multi-scale feature self-attention network for lightweight medical image segmentation DOI

Jiahui Zhong,

Wenhong Tian, Yuanlun Xie

et al.

Computer Methods and Programs in Biomedicine, Journal Year: 2025, Volume and Issue: unknown, P. 108611 - 108611

Published: Jan. 1, 2025

Language: Английский

Citations

3
Leveraging U-Net and selective feature extraction for land cover classification using remote sensing imagery DOI Creative Commons
Leo Ramos, Ángel D. Sappa

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Jan. 4, 2025

In this study, we explore an enhancement to the U-Net architecture by integrating SK-ResNeXt as encoder for Land Cover Classification (LCC) tasks using Multispectral Imaging (MSI). introduces cardinality and adaptive kernel sizes, allowing better capture multi-scale features adjust more effectively variations in spatial resolution, thereby enhancing model's ability segment complex land cover types. We evaluate approach Five-Billion-Pixels dataset, composed of 150 large-scale RGB-NIR images over 5 billion labeled pixels across 24 categories. The achieves notable improvements baseline U-Net, with gains 5.312% Overall Accuracy (OA) 8.906% mean Intersection Union (mIoU) when RGB configuration. With RG-NIR configuration, these increase 6.928% OA 6.938% mIoU, while configuration yields 5.854% 7.794% mIoU. Furthermore, not only outperforms other well-established models such DeepLabV3, DeepLabV3+, Ma-Net, SegFormer, PSPNet, particularly but also surpasses recent state-of-the-art methods. Visual tests confirmed superiority, showing that studied certain classes, lakes, rivers, industrial areas, residential vegetation, where architectures struggled achieve accurate segmentation. These results demonstrate potential capability explored handle MSI enhance LCC results.

Language: Английский

Citations

2
PAMSNet: A medical image segmentation network based on spatial pyramid and attention mechanism DOI
Yuncong Feng, Xiaoyan Zhu, Xiaoli Zhang

et al.

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 94, P. 106285 - 106285

Published: April 1, 2024

Language: Английский

Citations

8
ResDAC-Net: a novel pancreas segmentation model utilizing residual double asymmetric spatial kernels DOI Creative Commons
Zhanlin Ji, Jianuo Liu,

Juncheng Mu

et al.

Medical & Biological Engineering & Computing, Journal Year: 2024, Volume and Issue: 62(7), P. 2087 - 2100

Published: March 8, 2024

Abstract The pancreas not only is situated in a complex abdominal background but also surrounded by other organs and adipose tissue, resulting blurred organ boundaries. Accurate segmentation of pancreatic tissue crucial for computer-aided diagnosis systems, as it can be used surgical planning, navigation, assessment organs. In the light this, current paper proposes novel Residual Double Asymmetric Convolution Network (ResDAC-Net) model. Firstly, newly designed ResDAC blocks are to highlight features. Secondly, feature fusion between adjacent encoding layers fully utilizes low-level deep-level features extracted blocks. Finally, parallel dilated convolutions employed increase receptive field capture multiscale spatial information. ResDAC-Net highly compatible existing state-of-the-art models, according three (out four) evaluation metrics, including two main ones performance (i.e., DSC Jaccard index). Graphical abstract

Language: Английский

Citations

6
GLAC-Unet: Global-Local Active Contour Loss with an Efficient U-Shaped Architecture for Multiclass Medical Image Segmentation DOI
Minh-Nhat Trinh, Thi-Thao Tran, Do-Hai-Ninh Nham

et al.

Deleted Journal, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 17, 2025

The field of medical image segmentation powered by deep learning has recently received substantial attention, with a significant focus on developing novel architectures and designing effective loss functions. Traditional functions, such as Dice Cross-Entropy loss, predominantly rely global metrics to compare predictions labels. However, these measures often struggle address challenges occlusion nonuni-form intensity. To overcome issues, in this study, we propose function, termed Global-Local Active Contour (GLAC) which integrates both local features, reformulated within the Mumford-Shah framework extended for multiclass segmentation. This approach enables neural network model be trained end-to-end while simultaneously segmenting multiple classes. In addition this, enhance U-Net architecture incorporating Dense Layers, Convolutional Block Attention Modules, DropBlock. These improvements enable more effectively combine contextual information across layers, capture richer semantic details, mitigate overfitting, resulting precise outcomes. We validate our proposed method, namely GLAC-Unet, utilizes GLAC conjunction modified U-shaped architecture, three biomedical datasets that span range modalities, including two-dimensional three-dimensional images, dermoscopy, cardiac magnetic resonance imaging, brain imaging. Extensive experiments demonstrate promising performance approach, achieving score (DSC) 0.9125 ISIC-2018 dataset, 0.9260 Automated Cardiac Diagnosis Challenge (ACDC) 2017, 0.927 Infant Brain MRI Segmentation 2019. Furthermore, statistical significance testing p-values consistently smaller than 0.05 ACDC confirms superior method compared other state-of-the-art models. results highlight robustness effectiveness technique, underscoring its potential analysis. Our code will made available at https://github.com/minhnhattrinh312/Active-Contour-Loss-based-on-Global-and-Local-Intensity.

Language: Английский

Citations

0
Automated detection of traumatic bleeding in CT images using 3D U-Net# and multi-organ segmentation DOI
Rizki Nurfauzi,

Ayaka Baba,

Taka‐aki Nakada

et al.

Biomedical Physics & Engineering Express, Journal Year: 2025, Volume and Issue: 11(2), P. 025026 - 025026

Published: Jan. 24, 2025

Traumatic injury remains a leading cause of death worldwide, with traumatic bleeding being one its most critical and fatal consequences. The use whole-body computed tomography (WBCT) in trauma management has rapidly expanded. However, interpreting WBCT images within the limited time available before treatment is particularly challenging for acute care physicians. Our group previously developed an automated detection method images. further reduction false positives (FPs) necessary clinical application. To address this issue, we propose novel CT using deep learning multi-organ segmentation; Methods: proposed integrates three-dimensional U-Net# model FP approach based on segmentation. segmentation targets bone, kidney, vascular regions, where FPs are primarily found during process. We evaluated dataset delayed-phase contrast-enhanced collected from four institutions; Results: detected 70.0% bleedings 76.2 FPs/case. processing our was 6.3 ± 1.4 min. Compared previous ap-proach, significantly reduced number while maintaining sensitivity.

Language: Английский

Citations

0
MorNet: Asymmetric UNet-like Network with Morphological Opening and Closing for Image Segmentation DOI
Meng Li, Juntong Yun, Jiang Du

et al.

Lecture notes in computer science, Journal Year: 2025, Volume and Issue: unknown, P. 232 - 246

Published: Jan. 1, 2025

Language: Английский

Citations

0
Fill-UNet: Extended Composite Semantic Segmentation DOI
Qunpo Liu, Yi Zhao, Weiping Ding

et al.

Applied Soft Computing, Journal Year: 2025, Volume and Issue: unknown, P. 112891 - 112891

Published: Feb. 1, 2025

Language: Английский

Citations

0
Bilateral-Aware and Multi-Scale Region Guided U-Net for precise breast lesion segmentation in ultrasound images DOI
Yangyang Li,

Xintong Hou,

Xuanting Hao

et al.

Neurocomputing, Journal Year: 2025, Volume and Issue: unknown, P. 129775 - 129775

Published: Feb. 1, 2025

Language: Английский

Citations

0
Landscape of 2D Deep Learning Segmentation Networks Applied to CT Scan from Lung Cancer Patients: A Systematic Review DOI
Somayeh Sadat Mehrnia,

Zhino Safahi,

Amin Mousavi

et al.

Deleted Journal, Journal Year: 2025, Volume and Issue: unknown

Published: March 4, 2025

The increasing rates of lung cancer emphasize the need for early detection through computed tomography (CT) scans, enhanced by deep learning (DL) to improve diagnosis, treatment, and patient survival. This review examines current prospective applications 2D- DL networks in CT segmentation, summarizing research, highlighting essential concepts gaps; Methods: Following Preferred Reporting Items Systematic Reviews Meta-Analysis guidelines, a systematic search peer-reviewed studies from 01/2020 12/2024 on data-driven population segmentation using structured data was conducted across databases like Google Scholar, PubMed, Science Direct, IEEE (Institute Electrical Electronics Engineers) ACM (Association Computing Machinery) library. 124 met inclusion criteria were analyzed. LIDC-LIDR dataset most frequently used; finding particularly relies supervised with labeled data. UNet model its variants used models medical image achieving Dice Similarity Coefficients (DSC) up 0.9999. reviewed primarily exhibit significant gaps addressing class imbalances (67%), underuse cross-validation (21%), poor stability evaluations (3%). Additionally, 88% failed address missing data, generalizability concerns only discussed 34% cases. emphasizes importance Convolutional Neural Networks, UNet, analysis advocates combined 2D/3D modeling approach. It also highlights larger, diverse datasets exploration semi-supervised unsupervised enhance automated diagnosis detection.

Language: Английский

Citations

0