Article
Computer Science, Information Systems
Petra Takacs, Levente Kovacs, Andrea Manno-Kovacs
Summary: This study introduces an improved brain tumor segmentation method utilizing visual saliency features on MRI image volumes. The novel approach combines deep learning techniques with handcrafted feature models, demonstrating enhanced segmentation performance through the use of healthy templates in the training process and fusion of saliency maps with convolutional neural networks' prediction maps.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Dongjing Shan, Xiongwei Zhang, Tieyong Cao, Limin Wang, Chao Zhang
Summary: In this article, a three-stage hierarchical neural network is proposed for saliency detection, combining fast R-CNN, self-attention mechanism, and global regression model. Experimental results demonstrate excellent performance on several benchmark datasets and comparisons with 12 previous methods were conducted.
IEEE INTELLIGENT SYSTEMS
(2021)
Article
Computer Science, Information Systems
Guixian Wu, Yuancheng Li
Summary: A new convolutional neural network architecture named CliqueNet is introduced for improving semantic segmentation, along with a new fully convolutional network called CyclicNet. By adding long skip connections and short skip connections, the issue of vanishing gradient can be effectively avoided in the network.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Chemistry, Analytical
Zhengyu Xia, Joohee Kim
Summary: Mask2Former, a transformer-based semantic segmentation method, suffers from poor performance in obtaining local features and segmenting small objects. To address this, we propose a simple yet effective architecture that introduces auxiliary branches to capture dense local features during training, leading to improved performance in learning local information and segmenting small objects. Experimental results demonstrate that our model achieves state-of-the-art performance (57.6% mIoU on ADE20K, 84.8% mIoU on Cityscapes datasets).
Article
Chemistry, Analytical
Marin Bencevic, Yuming Qiu, Irena Galic, Aleksandra Pizurica
Summary: Medical images are often too large for training machine learning models, so downsampling is commonly used but leads to loss of information. We propose a general approach called Segment-then-Segment for training neural networks on smaller input sizes using image crops. One network performs initial segmentation on a downscaled image, and another network segments the most significant crops of the full-resolution image using the first segmentation. Finally, the segmentation masks of each crop are combined to form the output image. We evaluated this approach on different medical image modalities and found significant improvements in segmentation performance compared to baseline models trained on downscaled images, especially in terms of pixel-wise recall.
Article
Plant Sciences
Xudong Li, Yuhong Zhou, Jingyan Liu, Linbai Wang, Jun Zhang, Xiaofei Fan
Summary: This study developed an integrated framework using deep learning and convolutional neural networks for the rapid and accurate detection of potato leaf diseases. The experimental results showed that the framework performed well in the segmentation, classification, and semantic segmentation of potato leaves, providing a new model approach for the identification and detection of potato diseases.
FRONTIERS IN PLANT SCIENCE
(2022)
Article
Engineering, Multidisciplinary
Haojie Guo, Dedong Yang
Summary: This paper introduces an improved semantic segmentation model named PRDNet, which utilizes ResNet and dilated convolution to simultaneously extract multi-layer features of medical images. The multi-layer features are fused according to the structure of feature pyramid network in the decoding stage. After experiments on CHAOS and ISIC2017 datasets, the proposed algorithm shows a 1%-4% improvement in different evaluation metrics compared to other algorithms.
Article
Computer Science, Artificial Intelligence
Yiheng Zhang, Ting Yao, Zhaofan Qiu, Tao Mei
Summary: This paper thoroughly analyzes the design of convolutional blocks and the ways of interactions across multiple scales from a lightweight standpoint for semantic segmentation. Based on the analysis, the authors propose Lightweight and Progressively-Scalable Networks (LPS-Net) that outperform several efficient semantic segmentation methods.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2023)
Article
Biochemical Research Methods
Yi Ding, Xue Qin, Mingfeng Zhang, Ji Geng, Dajiang Chen, Fuhu Deng, Chunhe Song
Summary: In this paper, we propose an image segmentation network called RLSegNet, which translates the image segmentation process into a series of decision-making problems using reinforcement learning. RLSegNet is a U-shaped network composed of three components: a feature extraction network, a Mask Prediction Network (MPNet), and an up-sampling network with a cascade attention module. Experimental results demonstrate that the proposed method achieves better segmentation performance in brain tumor segmentation.
IEEE-ACM TRANSACTIONS ON COMPUTATIONAL BIOLOGY AND BIOINFORMATICS
(2023)
Article
Environmental Sciences
Shiyan Pang, Xinyu Li, Jia Chen, Zhiqi Zuo, Xiangyun Hu
Summary: High-resolution remote sensing image change detection technology plays an important role in various applications such as land cover/use monitoring, natural disaster monitoring, and military target analysis. However, current deep learning-based methods have limitations in terms of dataset dependence and cross-domain generalization. To address this, a prior semantic information-guided change detection framework (PSI-CD) is proposed, which reduces the sample size of change detection by utilizing prior semantic information.
Article
Engineering, Electrical & Electronic
Zenan Shi, Haipeng Chen, Dong Zhang
Summary: This paper proposes a neural network model called TANet for image manipulation localization, which addresses the challenges of capturing subtle manipulation artifacts and coarse boundaries. By introducing a multi-scale transformer branch and an operator induction module, TANet outperforms existing methods in terms of accuracy and boundary localization.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Information Systems
Xiaofeng Ding, Tieyong Zeng, Jian Tang, Zhengping Che, Yaxin Peng
Summary: This paper proposes a novel semantic representation (SR) module for extracting semantic information in semantic segmentation tasks. The module enhances the representation ability of semantic context by utilizing global semantic information and improves the consistency of intraclass features by aggregating global features. Additionally, the SR module can be extended to build a semantic representation refinement network for enhancing the structural reasoning of the model through multiple-scale iterations.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Artificial Intelligence
Mohamed Chala, Benayad Nsiri, My Hachem El Yousfi Alaoui, Abdelmajid Soulaymani, Abdelrhani Mokhtari, Brahim Benaji
Summary: This paper presents an automatic method for blood vessel segmentation in retina images using deep Convolutional Neural Networks (CNN). The proposed multi-encoder decoder architecture shows promising results in terms of specificity, accuracy, and precision. Our method outperforms other CNN-based approaches, with a high precision rate on the DRIVE dataset.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Radiology, Nuclear Medicine & Medical Imaging
Surender Singh Samant, Arun Chauhan, Jagadish Dn, Vijay Singh
Summary: Digital pathology plays a vital role in accurately diagnosing kidneys for transplantation and identifying kidney diseases. Glomerulus detection in kidney tissue segments is a key challenge in kidney diagnosis. In this study, a deep learning-based method using convolutional neural networks is proposed for glomerulus detection from digitized kidney slide segments. Various networks including ResNets, UNet, LinkNet, and EfficientNet are employed to train the models. Experimental results on the NIH HuBMAP kidney whole slide image dataset show that the proposed method achieves the highest scores with a Dice coefficient of 0.942.
JOURNAL OF DIGITAL IMAGING
(2023)
Article
Computer Science, Artificial Intelligence
Zhan Li, Chunxia Zhang, Yongqin Zhang, Xiaofeng Wang, Xiaolong Ma, Hai Zhang, Songdi Wu
Summary: Brain tissue segmentation is improved by integrating 2D and 3D data using Context-assisted full Attention Network (CAN) to address the issues of high time complexity and low segmentation accuracy in brain MRI.
MEDICAL IMAGE ANALYSIS
(2023)
Article
Computer Science, Artificial Intelligence
Mahdi Ahmadi, Alireza Norouzi, Nader Karimi, Shadrokh Samavi, Ali Emami
EXPERT SYSTEMS WITH APPLICATIONS
(2020)
Article
Computer Science, Information Systems
Hamidreza Zarrabi, Ali Emami, Pejman Khadivi, Nader Karimi, Shadrokh Samavi
MULTIMEDIA TOOLS AND APPLICATIONS
(2020)
Article
Engineering, Biomedical
M. Hajabdollahi, R. Esfandiarpoor, P. Khadivi, S. M. R. Soroushmehr, N. Karimi, S. Samavi
COMPUTERIZED MEDICAL IMAGING AND GRAPHICS
(2020)
Article
Engineering, Biomedical
Zahra Nabizadeh-Shahre-Babak, Nader Karimi, Pejman Khadivi, Roshanak Roshandel, Ali Emami, Shadrokh Samavi
Summary: This paper proposes an approach using the bag of visual words and a neural network classifier to classify X-ray chest images into COVID-19 and non-COVID-19 with high performance. Experimental results show that extracting features with the bag of visual words leads to better classification accuracy compared to state-of-the-art techniques.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2021)
Article
Computer Science, Artificial Intelligence
Sajjad Abbasi, Mohsen Hajabdollahi, Pejman Khadivi, Nader Karimi, Roshanak Roshandel, Shahram Shirani, Shadrokh Samavi
Summary: Machine Learning and Artificial Intelligence have shown potential as diagnostic tools in the healthcare domain, but their usefulness is hindered by a lack of labeled data. Transfer Learning and Knowledge Distillation are partial solutions to this issue, but still have limitations that need to be considered.
ARTIFICIAL INTELLIGENCE IN MEDICINE
(2021)
Article
Computer Science, Information Systems
Maryam Karimi, Mansour Nejati, Pejman Khadivi, Nader Karimi, Shadrokh Samavi
Summary: The study proposed a fast-efficient algorithm for blind quality assessment of stereoscopic images. It utilized supervised dictionary learning to learn discriminative distortion-specific bases from structural features of stereoscopic images. In experiments, the method achieved an overall correlation of 97% with subjective scores on common datasets.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Engineering, Biomedical
Ghazaleh Ghorbanzadeh, Zahra Nabizadeh, Nader Karimi, Pejman Khadivi, Ali Emami, Shadrokh Samavi
Summary: This article presents a channel selection method combining a sequential search method with a genetic algorithm to improve the performance and efficiency of brain-computer interface systems.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Engineering, Biomedical
Zahra Sobhaninia, Nader Karimi, Pejman Khadivi, Shadrokh Samavi
Summary: This paper presents an approach that simultaneously segments and classifies brain tumors in MRI images using a framework that contains MRI image enhancement and tumor region detection. A network called Multiscale Cascaded Multitask Network is proposed, which is based on a multitask learning approach containing segmentation and classification tasks. The proposed method achieves high accuracy in both segmentation (96.27 and 95.88 for DCS and mean IoU, respectively) and classification (97.988 accuracy).
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Engineering, Biomedical
Mohammadreza Naderi, Nader Karimi, Ali Emami, Shahram Shirani, Shadrokh Samavi
Summary: This study aims to improve the performance of conditional Generative Adversarial Networks (cGANs) in translating images by learning the target domain distribution from limited data with the help of noise input. The proposed method achieves better model generalization and comparable results compared to state-of-the-art methods.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Computer Science, Information Systems
Mohammad Reza Naderi, Mohammad Hossein Givkashi, Nader Karimi, Shahram Shirani, Shadrokh Samavi
Summary: This study proposes an image retargeting method that aims to change the size of images while preserving important content and minimizing distortions. By utilizing techniques such as inpainting, seam carving, super-resolution, and optimization, the proposed method demonstrates commendable performance in improving visual quality.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Proceedings Paper
Engineering, Electrical & Electronic
Maedeh Jamali, Nader Karimi, Shadrokh Samavi
Summary: The traditional visual quality evaluation criteria such as PSNR and MSE lack appropriate standards based on the human visual system. A weighted fuzzy-based criterion proposed in this paper aims to find essential parts of an image based on the HVS for more accurate evaluation of image quality. Experimental results demonstrate significant improvements over standard PSNR evaluations.
2021 29TH IRANIAN CONFERENCE ON ELECTRICAL ENGINEERING (ICEE)
(2021)