Article
Computer Science, Artificial Intelligence
Narinder Singh Punn, Sonali Agarwal
Summary: This article presents the application of U-Net architecture in biomedical image segmentation and provides a comprehensive analysis of U-Net variants. It highlights the success of these approaches in the ongoing pandemic and other areas, while also discussing the challenges and future research directions in biomedical image segmentation.
ARTIFICIAL INTELLIGENCE REVIEW
(2022)
Article
Biology
Hasib Zunair, A. Ben Hamza
Summary: Sharp U-Net architecture addresses the issues of blurred feature maps and over-/under-segmented target regions by employing a depthwise convolution before merging encoder and decoder features. The model consistently outperforms recent baselines in both binary and multi-class biomedical image segmentation tasks, while adding no extra learnable parameters and surpassing baselines with more than three times the number of parameters.
COMPUTERS IN BIOLOGY AND MEDICINE
(2021)
Article
Computer Science, Artificial Intelligence
Shukai Yang, Xiaoqian Zhang, Yufeng Chen, Youtao Jiang, Quan Feng, Lei Pu, Feng Sun
Summary: In recent years, there has been much attention on precise medical image segmentation methods based on the encoder-decoder structure. However, there are still limitations, such as increasingly complex network structures and insufficient multiscale information fusion ability. To address these issues, a novel lightweight precise medical image segmentation network called UcUNet was designed, which has a large receptive field and multiscale information fusion ability with a low parameter count.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Jihong Ouyang, Siguang Liu, Hao Peng, Harish Garg, Dang N. H. Thanh
Summary: This study proposes an improved U-Net based model for segmenting retinal vessel images. By introducing a local feature enhancement module and attention mechanism, the feature extraction of small vessels and vessel segmentation information is enhanced, improving the accuracy of medical diagnosis. The experiments on the DRIVE dataset demonstrate the potential clinical applications of the proposed method.
COMPLEX & INTELLIGENT SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Gang Sha, Junsheng Wu, Bin Yu
Summary: In this paper, a scheme for spinal fracture lesions segmentation based on U-net was proposed, which introduces an attention module and dilated convolution to achieve more accurate lesions segmentation. The attention module focuses on specific regions to improve the model's recognition of lesions, while dilated convolution increases the receptive field for more lesion feature information. Experimental results show that the proposed network outperforms U-net in lesions segmentation performance.
NEURAL PROCESSING LETTERS
(2021)
Article
Chemistry, Multidisciplinary
Sunetra Banerjee, Juan Lyu, Zixun Huang, Hung Fat Frank Leung, Timothy Tin-Yan Lee, De Yang, Steven Su, Yongping Zheng, Sai-Ho Ling
Summary: Scoliosis is a widespread medical condition mainly affecting young adults, necessitating periodic assessment for early detection. A non-radiating 3D ultrasound imaging technique has been developed as a safe alternative, but issues with low-contrast images and speckle noise require skilled intervention. To address the limitations of human expert interventions, an automatic, fast, and low-computation assessment technique is being developed for mass scoliosis diagnosis using a novel hybridized convolutional neural network architecture called Light-convolution Dense Selection U-Net (LDS U-Net). Testing with a dataset of 109 spine ultrasound images shows that LDS U-Net outperforms other models in terms of segmentation performance while requiring fewer parameters and less memory.
APPLIED SCIENCES-BASEL
(2021)
Article
Chemistry, Analytical
Lingyun Li, Hongbing Ma
Summary: The proposed hybrid variable structure-RDCTrans U-Net for liver tumor segmentation combines ResNeXt50, dilated convolution, and Transformer, achieving state-of-the-art results on the LiTS dataset with high segmentation accuracy.
Article
Biology
Shirsha Bose, Ritesh Sur Chowdhury, Rangan Das, Ujjwal Maulik
Summary: Biomedical image segmentation is crucial for medical image analysis, and deep learning algorithms allow for the design of advanced models to solve segmentation problems. The D3MSU-Net is proposed, which varies the receptive field at each level based on the resolution layer's depth and performs supervision at each resolution level. Experimental results demonstrate the superiority of the proposed network.
COMPUTERS IN BIOLOGY AND MEDICINE
(2022)
Article
Computer Science, Artificial Intelligence
Hengxin Liu, Guoqiang Huo, Qiang Li, Xin Guan, Ming-Lang Tseng
Summary: This study proposes a lightweight automatic 3D algorithm with an attention mechanism for brain-tumor image segmentation. By replacing standard convolutions, adding dilated convolutions, and introducing an attention mechanism, the proposed model achieves good performance in segmentation precision and parameter efficiency.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Vatsala Anand, Sheifali Gupta, Deepika Koundal, Karamjeet Singh
Summary: Skin, as a vital organ, plays a crucial role in providing protection to the human body from external surroundings. Early detection of skin diseases is necessary to improve survival rates and prevent the development of skin cancer. However, diagnosing skin diseases at an early stage, which can increase life expectancy, is challenging due to their similar appearance. To address this issue, a new innovative automated system combining the U-Net and Convolution Neural Network models is proposed for quick and accurate identification of skin lesions. The proposed model achieves an accuracy of 97.96% using the Adadelta optimizer on the HAM10000 dataset, which contains 10,015 dermoscopy images of seven different classifications of skin diseases.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Chemistry, Multidisciplinary
Yan Zhang, Kefeng Li, Guangyuan Zhang, Zhenfang Zhu, Peng Wang
Summary: In the field of computer vision, image segmentation plays a crucial role in addressing issues such as rapid changes in high-speed railroad image scenes, low segmentation accuracy, and significant information loss. We propose DFA-UNet, a segmentation algorithm based on an improved U-Net network architecture. By embedding the DFA attention module in the encoder part of the model, we enable efficient feature extraction and integration of channel feature weights. Evaluation on the RailSem19 dataset demonstrates our model's superior performance, achieving improvements in mIoU, F1-score, Accuracy, Precision, and Recall compared to U-Net.
APPLIED SCIENCES-BASEL
(2023)
Article
Oncology
Josef A. Buchner, Jan C. Peeken, Lucas Etzel, Ivan Ezhov, Michael Mayinger, Sebastian M. Christ, Thomas B. Brunner, Andrea Wittig, Bjoern H. Menze, Claus Zimmer, Bernhard Meyer, Matthias Guckenberger, Nicolaus Andratschke, Rami A. El Shafie, Juergen Debus, Susanne Rogers, Oliver Riesterer, Katrin Schulze, Horst J. Feldmann, Oliver Blanck, Constantinos Zamboglou, Konstantinos Ferentinos, Angelika Bilger, Anca Grosu, Robert Wolff, Jan S. Kirschke, Kerstin Eitz, Stephanie Combs, Denise Bernhardt, Daniel Rueckert, Marie Piraud, Benedikt Wiestler, Florian Kofler
Summary: This study compares the performance of different MRI sequences in automated brain metastasis (BM) segmentation. The T1-CE sequence alone achieved the best segmentation performance for BM, while the combination of T1-CE and T2-FLAIR sequences was important for edema segmentation.
RADIOTHERAPY AND ONCOLOGY
(2023)
Article
Radiology, Nuclear Medicine & Medical Imaging
S. Saumiya, S. Wilfred Franklin
Summary: In this study, the RDSDSU-Net model is proposed to enhance the accuracy of liver and liver tumour segmentation by utilizing residual deformable convolution and a convolutional spatial and channel features split graph network. Experimental results demonstrate that the proposed method achieves better segmentation results compared to existing techniques.
JOURNAL OF DIGITAL IMAGING
(2023)
Article
Engineering, Biomedical
Yu Yan, Yangyang Liu, Yiyun Wu, Hong Zhang, Yameng Zhang, Lin Meng
Summary: In this paper, an Attention Enhanced U-net with hybrid dilated convolution (AE U-net with HDC) model was proposed to segment breast tumors in ultrasound images, achieving higher accuracy and IOU values through the addition of a new loss function and integration of HDC modules.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2022)
Article
Public, Environmental & Occupational Health
Yugen Yi, Changlu Guo, Yangtao Hu, Wei Zhou, Wenle Wang
Summary: In this study, a novel network model called BCR-UNet is proposed, which effectively preserves tiny blood vessels in low-contrast peripheral regions and outperforms previous state-of-the-art methods on four public datasets.
FRONTIERS IN PUBLIC HEALTH
(2022)
Article
Multidisciplinary Sciences
Nematullo Rahmatov, Faisal Saeed, Anand Paul
Summary: This study estimates the vulnerability of AS routers and predicts AS neighbor values using ICMP packets and LSR approach. The results show a potential threat of ICMP flood attack to many ASs and a lack of firewall system deployment in some ASs at their network boundaries. It is predicted that AS neighbor numbers will reduce in the next 3 years.
SCIENTIFIC REPORTS
(2022)
Article
Multidisciplinary Sciences
Barathi Subramanian, Bekhzod Olimov, Shraddha M. Naik, Sangchul Kim, Kil-Houm Park, Jeonghong Kim
Summary: Sign language recognition faces challenges in accurate tracking, occlusion, and computational cost. To overcome these, an optimized MOPGRU model is proposed, achieving better prediction, efficiency, and information processing than other sequential models.
SCIENTIFIC REPORTS
(2022)
Article
Computer Science, Interdisciplinary Applications
Bekhzod Alisher Ugli Olimov, Kalyana C. Veluvolu, Anand Paul, Jeonghong Kim
Summary: This paper proposes an unsupervised learning-based computationally inexpensive, efficient, and interpretable model UzADL for automated visual inspection (AVI). The system annotates unlabeled images using a pseudo-labeling algorithm and explicitly visualizes the defective regions of identified abnormal instances using an anomaly interpretation technique. Experimental results demonstrate that UzADL outperforms other methods in terms of accuracy and speed.
COMPUTERS & INDUSTRIAL ENGINEERING
(2022)
Article
Engineering, Multidisciplinary
Urfa Malik Gul, Anand Paul, K. -W. -A. Chee
Summary: Mathematical modeling is an implementation of mathematics in real-world problems with the aim of better understanding them. This paper discusses the principles and procedures of mathematical modeling using formulas and equations, investigates the suitability of different methods, and emphasizes the importance of mathematical modeling technologies in computational tools.
MATHEMATICAL PROBLEMS IN ENGINEERING
(2022)
Review
Computer Science, Software Engineering
Barathi Subramanian, Anand Paul, Jeonghong Kim, K. -W. -A. Chee
Summary: This article provides a survey on widely used distance metrics and the challenges associated with this field. It discusses the significance of distance metrics in improving machine learning and deep learning models, and focuses on the application and success of Siamese and triplet networks in deep metric learning. The article comprehensively examines and evaluates complex factors such as sampling strategy, suitable distance metric, and network structure, making it an important and valuable research contribution.
SCIENTIFIC PROGRAMMING
(2022)
Article
Computer Science, Artificial Intelligence
Arun Saco, P. Shanmuga Sundari, J. Karthikeyan, Anand Paul
Summary: Advanced technologies like machine learning and artificial intelligence can improve the performance of PEM fuel cells, thereby increasing the efficiency of renewable energy utilization. Experimental study on different humidification processes reveals that the LR model provides better accuracy compared to other models.
Review
Chemistry, Analytical
Oumaima Moutik, Hiba Sekkat, Smail Tigani, Abdellah Chehri, Rachid Saadane, Taha Ait Tchakoucht, Anand Paul
Summary: Understanding actions in videos is a significant challenge in computer vision, and research has been conducted on this topic for decades. Convolutional neural networks (CNNs) have played a crucial role and have been widely used in deep learning for visual data exploitation and various computer vision tasks, including action recognition. However, with the emergence of the Vision Transformer models (ViTs) and their success in natural language processing, there is a discussion on whether they will replace CNNs in action recognition in video clips. This study provides a detailed analysis of this trending topic, comparing CNNs and Transformers for action recognition and discussing the trade-off between accuracy and complexity.
Review
Environmental Sciences
Malik Urfa Gul, Anand Paul, S. Manimurugan, Abdellah Chehri
Summary: Hydrotropism is a plant's movement or growth towards water, triggered by the plant's ability to detect water through various stimuli. This study aims to provide an overview of root movement towards water and plant water uptake stabilization. Hydrotropism is important for plants to survive in water-scarce environments and to efficiently utilize nutrients in the soil.
Review
Chemistry, Physical
S. Divya, Swati Panda, Sugato Hajra, Rathinaraja Jeyaraj, Anand Paul, Sang Hyun Park, Hoe Joon Kim, Tae Hwan Oh
Summary: Recent advancements in AI and ML have increased the demand for self-powered devices. To address the energy issue, energy-harvesting technologies like PENG and TENG are being explored. This article discusses the use of AI technologies for data processing in PENG and TENG, and the potential applications in robotics, security systems, and healthcare. The challenges and alternatives in these technologies are also explored.
Article
Computer Science, Information Systems
Alfred Daniel John William, Santhosh Rajendran, Pradish Pranam, Yosuva Berry, Anuj Sreedharan, Junaid Gul, Anand Paul
Summary: Blockchain technology is utilized for managing and safeguarding digital interactions in decentralized systems. In the proposed framework for consumer electronics data sharing and secure payments, an IoT meter transmits monthly consumption data to a decentralized application stored in the blockchain, generating bills and incentivizing legitimate consumers.
Article
Multidisciplinary Sciences
Bekhzod Olimov, Barathi Subramanian, Rakhmonov Akhrorjon Akhmadjon Ugli, Jea-Soo Kim, Jeonghong Kim
Summary: Extracting useful features at multiple scales is crucial in computer vision. However, current state-of-the-art methods have limitations in computation efficiency and generalization on small-scale images. To address these problems, we propose a novel image classification system based on elaborate data preprocessing steps and a carefully designed CNN model architecture. Our implemented CMSFL-Net achieves faster training/inference, higher accuracy, and better efficiency-speed trade-off compared to existing efficient networks.
SCIENTIFIC REPORTS
(2023)
Proceedings Paper
Computer Science, Information Systems
Bekhzod Olimov, Barathi Subramanian, Jeonghong Kim
Summary: This paper introduces a deep learning-based unsupervised learning method for automatizing the visual inspection process. The proposed method utilizes a pseudo-labeling algorithm and a graph Laplacian matrix to convert the autoencoder problem into a classification task, resulting in fast and precise performance. In experiments, the proposed method outperforms currently available methods in terms of speed and accuracy.
2022 THIRTEENTH INTERNATIONAL CONFERENCE ON UBIQUITOUS AND FUTURE NETWORKS (ICUFN)
(2022)
Review
Computer Science, Information Systems
Tarana Singh, Arun Solanki, Sanjay Kumar Sharma, Anand Nayyar, Anand Paul
Summary: Smart City is an emerging research domain that has attracted the attention of government, businesses, and researchers. This research paper provides a systematic literature review of the smart city domain, discussing its origin, definitions, characteristics, and components. Through a comprehensive literature survey, the paper identifies challenges and future trends in the smart city field. It serves as a guide for government, businesses, and researchers aiming to enhance the concept of smart cities.
Article
Computer Science, Information Systems
Barathi Subramanian, Jeonghong Kim, Mohammed Maray, Anand Paul
Summary: Emotion recognition in healthcare has attracted significant attention due to advancements in machine learning and deep learning techniques. We developed a real-time emotion recognition system that utilizes a digital twin setup to provide personalized healthcare treatment. The system achieved promising results without compromising accuracy.