Article
Computer Science, Interdisciplinary Applications
Zhiyong Xiao, Yixin Su, Zhaohong Deng, Weidong Zhang
Summary: This study enhances the segmentation of MR images using a semi-supervised learning method with a dual-teacher structure, utilizing a small amount of labeled data and a large amount of unlabeled data. The method significantly improves the segmentation results of MR images with high accuracy.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2022)
Article
Biology
Xiao Liu, Hongyi Chen, Chong Yao, Rui Xiang, Kun Zhou, Peng Du, Weifan Liu, Jie Liu, Zekuan Yu
Summary: Image fusion techniques are widely used in multi-modal medical image fusion tasks. However, most existing methods neglect the textural details and contrast between the tissues in regions of interest, which can distort important tumor information and limit the clinical applicability of the fused images. To address this issue, we propose a multi-modal MRI fusion generative adversarial network (BTMF-GAN) that aims to achieve a balance between tissue details and structural contrasts in brain tumor, an important region for medical applications.
COMPUTERS IN BIOLOGY AND MEDICINE
(2023)
Article
Environmental Sciences
Baisen Liu, Yuanjia Liu, Wulin Zhang, Yiran Tian, Weili Kong
Summary: This paper proposes a spectral Swin Transformer network for hyperspectral image classification, which achieves high accuracy and good generalization ability on multiple public datasets. The approach provides a new direction for utilizing deep learning techniques in the field of hyperspectral imaging.
Article
Computer Science, Artificial Intelligence
Benteng Ma, Yu Feng, Geng Chen, Changyang Li, Yong Xia
Summary: Medical data sharing is crucial but suffers from privacy issues. This paper proposes a novel federated learning algorithm, FedAR, which addresses data heterogeneity by employing a flexible re-weighting scheme and achieves superior performance.
PATTERN RECOGNITION
(2023)
Article
Engineering, Biomedical
Shenhai Zheng, Jiaxin Tan, Chuangbo Jiang, Laquan Li
Summary: This study aims to design, propose, and validate a deep learning method that extends the application of Transformer to multi-modality medical image segmentation. A novel automated multi-modal Transformer network called AMTNet is introduced for 3D medical image segmentation, and comprehensive experimental analysis on the Prostate and BraTS2021 datasets demonstrates significant improvements over the state-of-the-art segmentation networks. This powerful network enriches the research of the Transformer to multi-modal medical image segmentation.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Junyu Chen, Eric C. Frey, Yufan He, William P. Segars, Ye Li, Yong Du
Summary: Convolutional neural networks (ConvNets) have been a major focus in medical image analysis, but their performance is limited by a lack of consideration for long-range spatial relationships in images. Vision Transformer architectures have recently been proposed to address this issue and have shown state-of-the-art performances in medical imaging applications. In this paper, the researchers propose a hybrid Transformer-ConvNet model called TransMorph for volumetric medical image registration. The proposed model improves the performance significantly compared to existing registration methods and Transformer architectures, demonstrating the effectiveness of Transformers for medical image registration.
MEDICAL IMAGE ANALYSIS
(2022)
Article
Neurosciences
Yang Xu, Xianyu He, Guofeng Xu, Guanqiu Qi, Kun Yu, Li Yin, Pan Yang, Yuehui Yin, Hao Chen
Summary: In the field of medical image segmentation, traditional solutions mainly adopt convolutional neural networks (CNNs). This paper proposes a hybrid feature extraction network that combines CNNs and Transformer to better utilize global information for feature extraction and improve the segmentation performance of medical images. Additionally, a multi-dimensional statistical feature extraction module is also introduced to enhance low-dimensional texture features and further improve the segmentation results.
FRONTIERS IN NEUROSCIENCE
(2022)
Article
Computer Science, Artificial Intelligence
Quan Zhou, Shaozhuang Ye, Mingwei Wen, Zhiwen Huang, Mingyue Ding, Xuming Zhang
Summary: This paper proposes an unsupervised multi-modal medical image fusion method based on deep learning and transformers. By using a densely-connected high-resolution network and a hybrid transformer, this method can effectively capture local information and long-range dependencies, resulting in better fused results.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Zhenbing Liu, Fengfeng Wu, Yumeng Wang, Mengyu Yang, Xipeng Pan
Summary: In this study, a federated contrastive learning (FedCL) approach is proposed for training a shared deep learning model with privacy protection in distributed medical institutions. By integrating the idea of contrastive learning into the federated learning framework, FedCL combines the local model and the global model for contrastive learning, improving the generalization ability of the model. Experimental results on two public datasets demonstrate that our method outperforms other federated learning algorithms in medical image classification.
PATTERN RECOGNITION
(2023)
Article
Engineering, Biomedical
Rafsanjany Kushol, Collin C. Luk, Avyarthana Dey, Michael Benatar, Hannah Briemberg, Annie Dionne, Nicolas Dupre, Richard Frayne, Angela Genge, Summer Gibson, Simon J. Graham, Lawrence Korngut, Peter Seres, Robert C. Welsh, Alan H. Wilman, Lorne Zinman, Sanjay Kalra, Yee-Hong Yang
Summary: This study introduces a framework called SF2Former to distinguish ALS subjects from the control group by leveraging the power of the vision transformer architecture and combining spatial and frequency domain information. The proposed architecture is extensively evaluated with multi-modal neuroimaging data and proves to have superior classification accuracy compared to popular deep learning-based techniques.
COMPUTERIZED MEDICAL IMAGING AND GRAPHICS
(2023)
Article
Engineering, Biomedical
Jing Wang, Haiyue Zhao, Wei Liang, Shuyu Wang, Yan Zhang
Summary: This study develops a deep learning approach based on a cross-convolutional transformer for multi-organ segmentation in various medical images. It achieves better generalization and accuracy by integrating local and global contexts and establishing reliable relational connections.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Geochemistry & Geophysics
Bo Zhang, Yaxiong Chen, Yi Rong, Shengwu Xiong, Xiaoqiang Lu
Summary: Based on previous work, we propose an HSI classification network called MATNet, which combines multi-attention and transformer. The network uses spatial attention and channel attention to focus on more important information parts, then utilizes a tokenizer module for semantic-level representation and a transformer encoder module for deep semantic feature extraction. We also design a loss function called Lpoly to accommodate different datasets and tasks. Experimental results demonstrate that MATNet performs well in extracting spatial-spectral features of HSIs and understanding semantic degrees.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Chemistry, Analytical
Hojun Lee, Hyunjun Cho, Jieun Park, Jinyeong Chae, Jihie Kim
Summary: Transformer-based approaches have shown good results in image captioning tasks, but they have limitations in generating text from global features of an entire image. Therefore, this study proposes novel methods using a Global-Local Visual Extractor (GLVE) and a Cross Encoder-Decoder Transformer (CEDT) to generate better image captioning.
Article
Computer Science, Artificial Intelligence
Hafiz Tayyab Mustafa, Pourya Shamsolmoali, Ik Hyun Lee
Summary: This study proposes an image fusion framework based on vision transformer and graph attention, which improves feature representation and texture recovery by utilizing patch repetition of the source images. Through evaluations on benchmark datasets, the proposed method demonstrates superior performance.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Geochemistry & Geophysics
Xuming Zhang, Yuanchao Su, Lianru Gao, Lorenzo Bruzzone, Xingfa Gu, Qingjiu Tian
Summary: This article proposes two types of lightweight self-attention modules (CLMSA and PLMSA) to reduce the memory and computation burden of the transformer model in hyperspectral image classification. A lightweight transformer (LiT) network is built with these modules, combining convolutional blocks and transformers to extract both local and long-range dependencies. Additionally, a controlled multiclass stratified (CMS) sampling strategy is used to generate appropriately sized training data and mitigate overfitting. Experimental results validate the effectiveness of the proposed design.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Computer Science, Information Systems
Yifan Gao, Lezhou Wu
Summary: This paper introduces a new method called NoGoZero+ to enhance the training process of AlphaZero and improve performance in a game similar to Go, NoGo. The method significantly speeds up training and achieves good competition results under limited resources.
Article
Medicine, General & Internal
Yin Dai, Yumeng Song, Weibin Liu, Wenhe Bai, Yifan Gao, Xinyang Dong, Wenbo Lv
Summary: This paper investigates the use of deep convolutional neural networks for multi-focus image fusion, merging MRI and PET neural images into multi-modal images to enhance the accuracy of PD image classification. The results show that the test accuracy rates of the multi-modal fusion dataset outperform those of the single-modal MRI dataset, indicating the effectiveness of the multi-focus image fusion method for PD image classification.
Proceedings Paper
Computer Science, Artificial Intelligence
Yifan Gao, Lezhou Wu, Haoyue Li
Summary: This paper introduces a novel positional attention-based UNet-style model (GomokuNet) for Gomoku AI, which combines positional information modules and multiscale features to improve the performance of zero learning networks. The quantitative results from ablation analysis show that GomokuNet outperforms previous state-of-the-art zero learning networks, demonstrating the potential to enhance zero learning efficiency and AI engine performance.
2021 IEEE CONFERENCE ON GAMES (COG)
(2021)