CXR-Seg: A Novel Deep Learning Network for Lung Segmentation from Chest X-Ray Images

Bioengineering, Journal Year: 2025, Volume and Issue: 12(2), P. 167 - 167

Published: Feb. 10, 2025

Over the past decade, deep learning techniques, particularly neural networks, have become essential in medical imaging for tasks like image detection, classification, and segmentation. These methods greatly enhanced diagnostic accuracy, enabling quicker identification more effective treatments. In chest X-ray analysis, however, challenges remain accurately segmenting classifying organs such as lungs, heart, diaphragm, sternum, clavicles, well detecting abnormalities thoracic cavity. Despite progress, these issues highlight need improved approaches to overcome segmentation difficulties enhance reliability. this context, we propose a novel architecture named CXR-Seg, tailored semantic of lungs from images. The proposed network mainly consists four components, including pre-trained EfficientNet an encoder extract feature encodings, spatial enhancement module embedded skip connection promote adjacent fusion, transformer attention at bottleneck layer, multi-scale fusion block decoder. performance CRX-Seg was evaluated on publicly available datasets (MC, Darwin, Shenzhen X-rays, TCIA brain flair MRI images). method achieved Jaccard index, Dice coefficient, sensitivity, specificity 95.63%, 97.76%, 98.77%, 98.00%, 99.05%on MC; 91.66%, 95.62%, 96.35%, 95.53%, 96.94% V7 Darwin COVID-19; 92.97%, 96.32%, 96.69%, 96.01%, 97.40% Tuberculosis CXR Dataset, respectively. Conclusively, offers comparison with state-of-the-art methods, better generalization

Language: Английский

---

ARCUNet: enhancing skin lesion segmentation with residual convolutions and attention mechanisms for improved accuracy and robustness

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

Published: March 18, 2025

Skin lesion segmentation presents significant challenges due to the high variability in size, shape, color, and texture presence of artifacts like hair, shadows, reflections, which complicate accurate boundary delineation. To address these challenges, we proposed ARCUNet, a semantic model including residual convolutions attention techniques improve accuracy skin segmentation, By incorporating mechanisms, ARCUNet enhances feature learning, stabilizes training, sharpens focus on boundaries for improved accuracy. Residual ensure better gradient flow faster convergence, while mechanisms refine selection by emphasizing critical regions suppressing irrelevant details. The was tested ISIC 2016, 2017, 2018 datasets with outstanding results measures 98.12%, 96.45%, 98.19%, Dice 94.68%, 91.21%, 95.34%, Jaccard 91.14%, 88.33%, 93.53%, respectively. These findings signify ability segment lesions accurately thus as an effective tool computerized disease diagnosis.

Language: Английский

---

Mamba- and ResNet-Based Dual-Branch Network for Ultrasound Thyroid Nodule Segmentation

Bioengineering, Journal Year: 2024, Volume and Issue: 11(10), P. 1047 - 1047

Published: Oct. 20, 2024

Accurate segmentation of thyroid nodules in ultrasound images is crucial for the diagnosis cancer and preoperative planning. However, challenging due to their irregular shape, blurred boundary, uneven echo texture. To address these challenges, a novel Mamba- ResNet-based dual-branch network (MRDB) proposed. Specifically, visual state space block (VSSB) from Mamba ResNet-34 are utilized construct dual encoder extracting global semantics local details, establishing multi-dimensional feature connections. Meanwhile, an upsampling-convolution strategy employed left decoder focusing on image size detail reconstruction. A convolution-upsampling used right emphasize gradual refinement recovery. facilitate interaction between details context within decoder, cross-skip connection introduced. Additionally, hybrid loss function proposed improve boundary performance nodules. Experimental results show that MRDB outperforms state-of-the-art approaches with DSC 90.02% 80.6% two public nodule datasets, TN3K TNUI-2021, respectively. Furthermore, experiments third external dataset, DDTI, demonstrate our method improves by 10.8% compared baseline exhibits good generalization clinical small-scale datasets. The can effectively accuracy has great potential applications.

Language: Английский

---

Hybrid Deep Learning Framework for Melanoma Diagnosis Using Dermoscopic Medical Images

Diagnostics, Journal Year: 2024, Volume and Issue: 14(19), P. 2242 - 2242

Published: Oct. 8, 2024

Melanoma, or skin cancer, is a dangerous form of cancer that the major cause demise thousands people around world.

Language: Английский

---

MicroSeg: Multi-scale fusion learning for microaneurysms segmentation

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 97, P. 106700 - 106700

Published: Aug. 8, 2024

Language: Английский

---