Article
Environmental Sciences
Li Yan, Xingfen Tang, Yi Zhang
Summary: In this paper, a GAN-based network named GSUGAN is proposed for improved DEM interpolation, outperforming traditional methods and CEDGAN both visually and quantitatively. The results show that gated convolution and symmetric dilated convolution structures perform slightly better, and GAN-based methods excel in visual quality, especially in complex terrains, compared to CNN-based methods.
Article
Biology
Meilin Liu, Zidong Wang, Han Li, Peishu Wu, Fuad E. Alsaadi, Nianyin Zeng
Summary: In this paper, a novel attention augmented Wasserstein generative adversarial network (AA-WGAN) is proposed for fundus retinal vessel segmentation. The proposed AA-WGAN can effectively handle the imperfect data property of segmenting tiny vessels, highlight regions of interests via attention augmented convolution, and suppress useless information through the squeeze-excitation module. The comprehensive evaluation on three datasets confirms the competitiveness of the proposed AA-WGAN, with accuracy of 96.51%, 97.19%, and 96.94% achieved on DRIVE, STARE, and CHASE_DB1 datasets respectively. The effectiveness of important components is validated by ablation study, demonstrating considerable generalization ability of the proposed AA-WGAN.
COMPUTERS IN BIOLOGY AND MEDICINE
(2023)
Article
Computer Science, Hardware & Architecture
Minqiang Yang, Yinru Ye, Kai Ye, Wei Zhou, Xiping Hu, Bin Hu
Summary: In this paper, a multi-scale generative adversarial network with class activation mapping is proposed to enhance the efficiency and accuracy of vessel segmentation using artificial intelligence. The incorporation of attention mechanism and multi-scale discrimination improves the ability to locate and segment fine retinal vessels and discriminate different receptive fields. The instability problem caused by unsupervised learning is addressed by introducing a supervised segmentation loss, and a data augmentation method is proposed for better generalization ability. Experimental results and comparisons with previous models demonstrate the superiority and effectiveness of the proposed model.
MOBILE NETWORKS & APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Yuchen Yuan, Lituan Wang, Lei Zhang
Summary: Despite achieving human-level performance in retinal vessel segmentation, deep learning based methods still lack vessel connectivity in the generated segmentation maps. To address this issue, a novel framework is proposed to enhance vessel connectivity by incorporating vessel structure into the segmentation network through adversarial learning. Experimental results on publicly available datasets demonstrate the efficacy of the proposed framework, which is independent of segmentation models and improves vessel connectivity without introducing extra memory or computational burden.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Biotechnology & Applied Microbiology
Shuang Xu, Zhiqiang Chen, Weiyi Cao, Feng Zhang, Bo Tao
Summary: The retinal vessel segmentation algorithm based on residual convolution neural network is able to accurately identify retinal vessels in fundus images, with excellent performance in achieving complete retinal vessel segmentation, connected vessel stems and terminals. The algorithm has proven to be effective in detecting more capillaries, with superior accuracy and specificity compared to other methods.
FRONTIERS IN BIOENGINEERING AND BIOTECHNOLOGY
(2021)
Article
Biology
Tabea Kossen, Pooja Subramaniam, Vince I. Madai, Anja Hennemuth, Kristian Hildebrand, Adam Hilbert, Jan Sobesky, Michelle Livne, Ivana Galinovic, Ahmed A. Khalil, Jochen B. Fiebach, Dietmar Frey
Summary: Anonymization and data sharing are essential for privacy protection and acquiring large datasets in medical image analysis, especially in neuroimaging. Generative adversarial networks (GANs) show potential in providing anonymous images while maintaining predictive properties. Among the three GANs tested, WGAN-GP-SN showed the highest performance in generating synthetic data for vessel segmentation with U-net. Transfer learning with synthetic data demonstrated improved model performance, particularly for individual patients.
COMPUTERS IN BIOLOGY AND MEDICINE
(2021)
Article
Computer Science, Interdisciplinary Applications
Haowen Pang, Shouliang Qi, Yanan Wu, Meihuan Wang, Chen Li, Yu Sun, Wei Qian, Guoyan Tang, Jiaxuan Xu, Zhenyu Liang, Rongchang Chen
Summary: In this study, two synthesizers were developed to achieve mutual synthesis between non-contrast CT (NCCT) and contrast-enhanced CT (CECT) using generative adversarial networks. The results demonstrated the effectiveness of the synthesizers in high-quality synthesis of NCCT and CECT images, with the training process being crucial to their performance.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2023)
Article
Computer Science, Information Systems
HaiYing Xia, LingYu Wu, Yang Lan, HaiSheng Li, ShuXiang Song
Summary: In this article, a hierarchical recurrent convolution neural network (HRNet) is proposed to improve the detection of weak retinal vessels. The HRNet integrates the advantages of ResNet and Squeeze and Excitation (SE), and employs a hierarchical recurrent mechanism to explore features from different layers.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Tariq M. Khan, Syed S. Naqvi, Antonio Robles-Kelly, Imran Razzak
Summary: This paper proposes a computer-aided diagnosis method for retinal diseases called Multi-resolution Contextual Network (MRC-Net). It learns contextual dependencies between semantically different features by extracting multi-scale features and uses bi-directional recurrent learning to model dependencies. Moreover, it improves the performance of the segmentation network through adversarial training for foreground segmentation. The method outperforms competitive approaches in terms of Dice score and Jaccard index on three benchmark datasets.
Article
Chemistry, Multidisciplinary
Yuepeng Zhou, Huiyou Chang, Yonghe Lu, Xili Lu
Summary: This study introduces CDTNet, an image classification model based on convolutional neural networks. CDTNet utilizes two branches with different dilation rates to capture multi-scale features and recovers low-resolution information through transposed convolution. Experimental results demonstrate that CDTNet outperforms state-of-the-art models on multiple benchmark datasets with lower loss, higher accuracy, and faster convergence speed.
APPLIED SCIENCES-BASEL
(2022)
Article
Engineering, Biomedical
Xiaojun Yu, Mingshuai Li, Chenkun Ge, Perry Ping Shum, Jinna Chen, Linbo Liu
Summary: In this study, a generative adversarial network with multi-scale convolution and dilated convolution res-network (MDR-GAN) is proposed to alleviate speckle noise in optical coherence tomography (OCT) imaging. Experimental results demonstrate that MDR-GAN outperforms other denoising methods.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Computer Science, Hardware & Architecture
Cong Bai, Anqi Zheng, Yuan Huang, Xiang Pan, Nan Chen
Summary: A framework using CNN and CGAN, as well as MGCN model, was proposed for image caption generation, which can better utilize visual relationships between objects and generate captions with semantic meanings.
Article
Computer Science, Software Engineering
Chen Jiqing, Wei Depeng, Long Teng, Luo Tian, Wang Huabin
Summary: This paper proposes an image enhancement network based on CycleGAN to improve road segmentation performance under severe weather conditions. By using an unsupervised CycleGAN network to enhance road image features and inputting the enhanced image into a semantic segmentation network, the segmentation of the drivable area of the road is achieved. Experimental results show that this method can significantly improve the performance of the original semantic segmentation network for road segmentation under severe weather conditions.
Article
Engineering, Mechanical
Yong Huang, Haoyu Zhang, Hui Li, Stephen Wu
Summary: This paper introduces a new CS method that replaces sparsity regularization with a generative model for automatic crack segmentation of compressed crack images. The method utilizes generative models to capture necessary features, demonstrating remarkable performance and energy savings.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2021)
Article
Computer Science, Information Systems
Binghua Xie, Hao Zhang, Cheolkon Jung
Summary: This paper proposes a weakly connected dense generative adversarial network, named WCDGAN, for artifacts removal of highly compressed images. Experimental results show that WCDGAN successfully removes compression artifacts and produces high-quality images.