Search

Numerous computer-aided leukemia detection methods have been introduced to overcome the limitations of clinical diagnosis procedures. The precision of computer-aided leukemia detection highly depends on the accurate segmentation of white blood cells (WBCs) from the stained whole slide image (WSI). This paper proposes WBC segmentation from WSI using U-Net and its variants. The major contributions to this paper are as follows. First, WBC segmentation is proposed using U-Net, U-Net++ and three transfer learning-based U-Net models, namely U-Net-VGG16, U-Net-VGG19, and U-Net-ResNet. Second, a comprehensive and comparative experimental analysis of the proposed WBC segmentation approaches is presented using four evaluation parameters, namely Dice coefficient, Intersection over Union (IoU), precision, and recall. Third, WBC segmentation approaches are also evaluated using four loss functions such as binary cross-entropy, focal, Dice, and IoU to identify the most effective loss function for WBC segmentation. Finally, the trained models' ability to perform WBC segmentation is validated to assess their practical applicability. The empirical results reveal that U-Net-VGG16 and U-Net-VGG19 achieve a high Dice coefficient of 90% and IoU of 83%. Besides, U-Net++ achieves high precision and recall of 98% and 70%, respectively. Although the results reveal that transfer learning-based U-Net models perform better, deployment of the trained model shows that the U-Net segments WBC more precisely than transfer learning-based U-Net models. [ABSTRACT FROM AUTHOR]

Anomaly detection is to identify abnormal events against normal ones within surveillance videos mainly collected in ground-based settings. Recently, the demand for processing drone-collected data is rapidly growing with the expanding range of drone applications. However, as most aerial videos collected by flying drones contain dynamic backgrounds and others, it is necessary to deal with their spatio-temporal features in detecting anomalies. This study presents a transformer-based video anomaly detection method whereby we investigate a challenging aerial dataset and three benchmark ground-based datasets. A multi-stage transformer is leveraged as an encoder to generate multi-scale feature maps, which are then conveyed to a hierarchical spatio-temporal transformer, that is linked to a decoder and used to capture spatial and temporal information by utilizing a joint attention mechanism. Extensive evaluations including several ablation studies suggest that this network outperforms the state-of-the-art methods. We expect the proposed transformer for U-net can find diverse applications in the video processing area. Code and pre-trained models are publicly available at https://github.com/vt-le/HSTforU. [ABSTRACT FROM AUTHOR]

In the aftermath of natural disasters, timely and accurate building damage assessment is crucial for effective disaster response. This research introduces an innovative integrated ResNet-U-Net ensemble model that significantly enhances the accuracy and efficiency of damage detection from satellite imagery. The model combines the deep feature extraction capabilities of ResNet with the precise localization advantages of U-Net, supplemented by attention mechanisms to focus on critical damage indicators. Methodologically, the research employs a dataset of high-resolution satellite images from post-Hurricane Iota, processed through normalization and augmentation techniques. The model demonstrates exceptional accuracy, achieving 96.91% on an unbalanced dataset representative of real-world conditions, and 89.35% on a balanced dataset. This performance indicates a substantial improvement over existing approaches, offering a rapid and reliable tool for aiding disaster relief efforts. The proposed model's computational efficiency and generalizability underscore its potential to transform disaster management practices. Future work will explore the model's application to various disaster types and integration with real-time data processing systems. [ABSTRACT FROM AUTHOR]

Objectives: To develop a novel automatic delineation model, the Multi-Scale Channel Attention U-Net (MCAU-Net) model, for gallbladder segmentation on CT images of patients with liver cancer. Methods: We retrospectively collected the CT images from 120 patients with liver cancer, based on which ground truth was manually delineated by physicians. The images and ground truth constitute a dataset, which was proportionally divided into a training set (54%), a validation set (6%), and a test set (40%). Data augmentation was performed on the training set. Our proposed MCAU-Net model was employed for gallbladder segmentation and its performance was evaluated using Dice Similarity Coefficient (DSC), Jaccard Similarity Coefficient (JSC), Positive Predictive Value (PPV), Sensitivity (SE), Hausdorff Distance (HD), Relative Volume Difference (RVD), and Volumetric Overlap Error (VOE) metrics. Results: On the test set, MCAU-Net achieved DSC, JSC, PPV, SE, HD, RVD, and VOE values of 0.85 ± 0.22, 0.79 ± 0.23, 0.92 ± 0.14, 0.84 ± 0.23, 2.75 ± 0.98, 0.18 ± 0.48, and 0.22 ± 0.42, respectively. Compared to the control models, U-Net, SEU-Net and TransUNet, the MCAU-Net improved DSC 0.06, 0.04 and 0.06, JSC by 0.09, 0.06 and 0.09, PPV by 0.08, 0.08 and 0.05, SE by 0.05,0.05 and 0.07, and reduced HD by 0.45, 0.28 and 0.41, RVD by 0.07, 0.03 and 0.07, VOE by 0.04, 0.02 and 0.08 respectively. Qualitative results revealed that MCAU-Net produced smoother and more accurate boundaries, closer to the expert delineation, with less over-segmentation and under-segmentation and improved robustness. Conclusions: The MCAU-Net model significantly improves gallbladder segmentation on CT images. It satisfies clinical requirements and enhances the efficiency of physicians, particularly in segmenting complex anatomical structures. [ABSTRACT FROM AUTHOR]

Ulnar variance is a relative length difference in the wrist between the ulna and radius bones. It is a critical factor in helping to diagnose wrist disorders. The typical standard classification of length difference (ulnar variance) is divided into three major types: positive ulnar variance, negative ulnar variance, and neutral ulnar variance. Conventional or manual methods of measuring ulnar variance are long and time-consuming. With the urgent need for high efficiency and high speed, achieving more accurate diagnoses has become essential. In this paper, a deep learning-based methodology is used to automatically detect ulnar variance from radiographic images. Advanced Convolutional Neural Networks are exploited instead of traditional manual methods. Specifically, U-Net is used in the segmentation of ulna and radius bones, while DenseNets are applied to classify the type of ulnar variance. The essential contribution of this work is collecting a dataset of fully annotated wrist radiographs that are specific to this topic, which can be used as a resource to train and validate our models. Another contribution of this paper is optimizing the DenseNets model's hyperparameters to enhance its performance. Our model achieved a segmentation accuracy of 97.7% and an ulna variance classification accuracy of 92.1%. It outperformed previous deep learning-based methods in automatically segmenting the ulna and radius. This advancement not only reduces diagnosis time but also improves result reliability. [ABSTRACT FROM AUTHOR]

Introduction: Magnetic resonance imaging (MRI) is conventionally used for the detection and diagnosis of multiple sclerosis (MS), often complemented by lumbar puncture—a highly invasive method—to validate the diagnosis. Additionally, MRI is periodically repeated to monitor disease progression and treatment efficacy. Recent research has focused on the application of artificial intelligence (AI) and radiomics in medical image processing, diagnosis, and treatment planning. Methods: A review of the current literature was conducted, analyzing the use of AI models and texture analysis for MS lesion segmentation and classification. The study emphasizes common models, including U-Net, Support Vector Machine, Random Forest, and K -Nearest Neighbors, alongside their evaluation metrics. Results: The analysis revealed a fragmented research landscape, with significant variation in model architectures and performance. Evaluation metrics such as Accuracy, Dice score, and Sensitivity are commonly employed, with some models demonstrating robustness across multi-center datasets. However, most studies lack validation in clinical scenarios. Discussion: The absence of consensus on the optimal model for MS lesion segmentation highlights the need for standardized methodologies and clinical validation. Future research should prioritize clinical trials to establish the real-world applicability of AI-driven decision support tools. This review provides a comprehensive overview of contemporary advancements in AI and radiomics for analyzing and monitoring emerging MS lesions in MRI. [ABSTRACT FROM AUTHOR]

Cloud classification is fundamental to advancing climate research and improving weather forecasting. However, existing cloud classification models are constrained by several limitations. For instance, simple statistical methods depend heavily on prior knowledge, leading to frequent misclassifications in regions with high latitudes or complex terrains. Machine learning approaches based on two-dimensional images face challenges such as data scarcity and high annotation costs, which hinder accurate pixel-level cloud identification. Additionally, single-pixel classification methods fail to effectively exploit the spatial correlations inherent in cloud structures. In this paper, we introduce the one-dimensional nested U-Net cloud-classification model (1D-CloudNet), which was developed using Himawari-8 and CloudSat data collected over two years (2016–2017), comprising a total of 27,688 samples. This model is explicitly tailored for the analysis of one-dimensional, multi-channel images. Experimental results indicate that 1D-CloudNet achieves an overall classification accuracy of 88.19% during the day and 87.40% at night. This represents a 3–4% improvement compared to traditional models. The model demonstrates robust performance for both daytime and nighttime applications, effectively addressing the absence of nighttime data in the Himawari-8 L2 product. In the future, 1D-CloudNet is expected to support regional climate research and extreme weather monitoring. Further optimization could enhance its adaptability to complex terrains. [ABSTRACT FROM AUTHOR]

The demand for large-scale topographic maps in Indonesia has significantly increased due to the implementation of several government initiatives that necessitate the utilization of spatial data in development planning. Currently, the national production capacity for large-scale topographic maps in Indonesia is 13,000 km2/year using stereo-plotting/mono-plotting methods from photogrammetric data, Lidar, high-resolution satellite imagery, or a combination of the three. In order to provide the necessary data to the respective applications in a timely manner, one strategy is to only generate critical layers of the maps. One of the topographic map layers that is often needed is land cover. This research focuses on providing land cover to support the accelerated provision of topographic maps. The data used are very-high-resolution satellite images. The method used is a deep learning approach to classify very-high-resolution satellite images into land cover data. The implementation of the deep learning approach can advance the production of topographic maps, particularly in the provision of land cover data. This significantly enhances the efficiency and effectiveness of producing large-scale topographic maps, hence increasing productivity. The quality assessment of this study demonstrates that the AI-assisted method is capable of accurately classifying land cover data from very-high-resolution images, as indicated by the Kappa values of 0.81 and overall accuracy of 86%, respectively. [ABSTRACT FROM AUTHOR]

Cardiovascular diseases (CVDs) are the primary cause of death worldwide. For accurate diagnosis of CVDs, robust and efficient ECG denoising is particularly critical in ambulatory cases where various artifacts can degrade the quality of the ECG signal. None of the present denoising methods preserve the morphology of ECG signals adequately for all noise types, especially at high noise levels. This study proposes a novel Fully-Gated Denoising Autoencoder (FGDAE) to significantly reduce the effects of different artifacts on ECG signals. The proposed FGDAE utilizes gating mechanisms in all its layers, including skip connections, and employs Self-organized Operational Neural Network (self-ONN) neurons in its encoder. Furthermore, a multi-component loss function is proposed to learn efficient latent representations of ECG signals and provide reliable denoising with maximal morphological preservation. The proposed model is trained and benchmarked on the QT Database (QTDB), degraded by adding randomly mixed artifacts collected from the MIT-BIH Noise Stress Test Database (NSTDB). The FGDAE showed the best performance on all seven error metrics used in our work in different noise intensities and artifact combinations compared with state-of-the-art algorithms. Moreover, FGDAE provides reliable denoising in extreme conditions and for varied noise compositions. The significantly reduced model size, 61% to 73% reduction, compared with the state-of-the-art algorithm, and the inference speed of the FGDAE model provide evident benefits in various practical applications. While our model performs best compared with other models tested in this study, more improvements are needed for optimal morphological preservation, especially in the presence of electrode motion artifacts. [ABSTRACT FROM AUTHOR]

Ocean Bottom Node (OBN) is a seismic data acquisition technique, comprising a hydrophone and a three-component geophone. In practice, the vertical component is susceptible to high-amplitude, low-velocity, and low-frequency shear wave noise, which negatively impacts the subsequent processing of dual-sensor data. The most commonly used method is adaptive matching subtraction, which estimates shear wave noise in the vertical component by solving an optimization problem. Neural networks, as robust nonlinear fitting tools, offer superior performance in resolving nonlinear mapping relationship and exhibit computational efficiency. In this paper, we introduce a self-supervised shear wave suppression approach for 3D OBN seismic data, using a neural network in place of the traditional adaptive matching subtraction operator. This method adopts the horizontal components as the input to the neural network, and measures the output and the noisy vertical component to establish a loss function for the network training. The network output is the predicted shear wave noise. To better balance signal leakage and noise suppression, the method incorporates a local normalized cross-correlation regularization term in the loss function. As a self-supervised method, it does not need clean data to serve as labels, thereby negating the tedious work of obtaining clean field data. Extensive experiments on both synthetic and field data demonstrate the effectiveness of the proposed method on shear wave noise suppression for 3D OBN seismic data. [ABSTRACT FROM AUTHOR]

Diabetic Retinopathy (DR) is a common and significant complication in patients with diabetes, and severely affecting their quality of life. Image segmentation plays a crucial role in the early diagnosis and treatment of DR. However, traditional methods are limited in terms of segmentation accuracy and generalization capability. This paper proposes a novel image segmentation method for diabetic retinopathy based on the integration of DenseNet and U-Net networks. Firstly, DenseNet is utilized to replace the encoder part of U-Net. The dense connection mechanism enhances the efficiency of feature propagation and improves the feature extraction capability of the encoder. Secondly, we introduce Omni-Dimensional Dynamic Convolution (ODConv) to replace traditional convolutions. ODConv is used to handle diverse input features effectively. The model's adaptability and segmentation accuracy for various samples is improved. Finally, we integrate the Convolutional Block Attention Module (CBAM) into the decoder. By leveraging channel and spatial attention, the ability to capture key features is enhanced, and the segmentation accuracy and boundary recognition capabilities are improved. To validate the effectiveness and feasibility of the model. This paper conducts experiments on the DDR dataset and the IDRID dataset. The experimental results show that the proposed method improved the mDice score by 0.53% compared to traditional methods. It also increased the mIoU score by 1.12% over traditional methods. This indicates that the proposed method has better performance in segmenting diabetic retinopathy images. It also demonstrates better generalization ability. [ABSTRACT FROM AUTHOR]

Composite laminates, known for their strength and flexibility, are widely used but can delaminate under certain barely visible impact. Ultrasound Transmission (UT) scans can detect such failures, but the noise in the scans makes automation of delamination morphology challenging. Machine learning could solve this, but it requires substantial training data. Furthermore, given the considerable time and cost of conducting impact tests on composite laminates, such data remains sparse. This study overcomes data sparsity by enlarging UT scan dataset using image augmentation techniques and synthetic data generation methods. A combination of rotation and elastic deformation of real UT images produced augmented data. Synthetic data was generated by mimicking the statistical variations in real data, including delamination length and rotation per ply, and adding gradients to match the original image noise. A U-Net based machine learning segmentation approach was utilized to capture local and global information through a series of convolutions to apply per-pixel classification. Systematic experimentation was conducted to study the influence of adding augmented and synthetic data on segmentation accuracy. The results show that the use of augmented data increases accuracy significantly. With the addition of synthetic data to the augmented data, there is a marginal improvement in accuracy. However, feature edge accuracy is significantly improved. The preliminary results presented here show the promising use of machine learning techniques for complex, non-destructive inspection methods. [ABSTRACT FROM AUTHOR]

Floods are among the most widespread and devastating natural disasters, accounting for 47% of all weather-related events and affecting over 2.3 billion people, particularly in Asia. Assessing flood-prone areas is crucial for effective disaster risk reduction, but existing flood damage estimation methods, such as depth-damage functions, often lack regional adaptability and accuracy. This study addresses this gap by integrating geospatial data, remote sensing, and artificial intelligence (AI) to identify flood-affected areas in the Kan basin, Tehran. We applied deep learning methods, specifically U-Net and fully convolutional neural network (FCN) algorithms, to optical and radar images from four flood events. Our results demonstrate that the U-Net model achieves significantly higher accuracy (88%) in identifying flood-affected areas compared to the FCN model (55% accuracy). This superior performance is further supported by the mean intersection over union (mIoU) values, with U-Net achieving 0.65, compared to 0.55 for FCN. The key message of this investigation is that deep learning, particularly the U-Net model, applied to remote sensing data holds significant promise for enhancing flood monitoring, early warning systems, and disaster management strategies by enabling more accurate and timely flood assessments. [ABSTRACT FROM AUTHOR]

Seismic data are fundamental for understanding subsurface geological structures and geological events. Previous studies have primarily focused on interpolating missing traces in seismic data, overlooking the reconstruction of complex geological events and patterns in 3D post-stack seismic images. A novel approach for the predictive reconstruction of missing geological events and patterns using our developed U-Net neural network model is proposed in this study. In contrast to existing methods that primarily interpolate traces, our proposed approach aims to predict and reconstruct complex missing geological events, including faults and unconformities, by leveraging the inherent patterns in seismic data. The impact of the different architectural choices, including various downsampling layers (max pooling, strided convolution, and dilated convolution) and loss functions such as SSIM, VGG19 perceptual loss, MSE, SNR, and PSNR, on the reconstruction process was comprehensively investigated in this study. The effectiveness of our proposed approach in generating realistic reconstructions of missing geological events and patterns was demonstrated by the experimental results. Specifically, the validation scores were found to be 0.86 for SSIM, 16.09 dB for SNR, and 16 dB for PSNR. Additionally, the validation losses were 0.0007 for VGG19 perceptual loss and 0.025 for MSE loss. These results underscore the ability of the model to capture the intricate geological and seismic features. The performance of the model in capturing the complexities of seismic data was further highlighted by a qualitative interpretation of the generated images. By addressing the limitations of existing methods and focusing on reconstructing geological events, our work advances the field of seismic data reconstruction. Our findings underscore the critical importance of a predictive reconstruction in seismic imaging and provide valuable insights for optimizing the architectural design choices. This study lays the groundwork for future research in this area, emphasizing the need for a nuanced understanding of architectural choices in seismic data reconstruction. [ABSTRACT FROM AUTHOR]

For non-destructive evaluation, the segmentation of infrared thermographic images of carbon fibre composites is a critical task in material characterisation and quality assessment. This paper presents a study on the application of image processing techniques, particularly adaptive thresholding, and advanced neural network models, including U-Net, DeepLabv3, and BiLSTM, for the segmentation of infrared images. This work introduces the innovative combination of DeepLabv3 and BiLSTM applied in infrared images of carbon fibre-reinforced polymer samples for the first time, proposing it as a novel approach for enhancing the accuracy of segmentation tasks. An experimental comparison of these models was conducted to assess their effectiveness in identifying artificial defects in these images. The performance of each model was evaluated using the F1-Score and Intersection over Union (IoU) metrics. The results demonstrate that the proposed combination of DeepLabv3 and BiLSTM outperforms other methods, achieving an F1-Score of 0.96 and an IoU of 0.83, showcasing its potential for advanced material analysis and quality control. [ABSTRACT FROM AUTHOR]

Raman spectroscopy is an extensively explored vibrational spectroscopic technique to analyze the biochemical composition and molecular structure of samples, which is often assumed to be non-destructive when carefully using proper laser power and exposure time. However, the inherently weak Raman signal and concurrent fluorescence interference often lead to Raman measurements with a low signal-to-noise ratio (SNR), especially for biological samples. Great efforts have been made to develop experimental approaches and/or numerical algorithms to improve the SNR. In this study, we proposed an ensemble learning approach to recover and denoise Raman measurements with a low SNR. The proposed ensemble learning approach was evaluated on 986 pairs of Raman measurements, each pair of which consists of a low SNR Raman spectrum and a high SNR reference Raman spectrum from the exact same fungal sample but uses 200 times the integration time. Compared with conventional methods, the Raman measurements recovered by the proposed ensemble learning approach are more identical to high SNR reference Raman measurements, with an average RMSE and MAE of only 1.337 × 10−2 and 1.066 × 10−2, respectively; thus, the proposed ensemble learning approach is expected to be a powerful tool for numerically improving the SNR of Raman measurements and further benefits rapid Raman acquisition from biological samples. [ABSTRACT FROM AUTHOR]

U-Net and its variants have been widely used in the field of image segmentation. In this paper, a lightweight multi-scale Ghost U-Net (MSGU-Net) network architecture is proposed. This can efficiently and quickly process image segmentation tasks while generating high-quality object masks for each object. The pyramid structure (SPP-Inception) module and ghost module are seamlessly integrated in a lightweight manner. Equipped with an efficient local attention (ELA) mechanism and an attention gate mechanism, they are designed to accurately identify the region of interest (ROI). The SPP-Inception module and ghost module work in tandem to effectively merge multi-scale information derived from low-level features, high-level features, and decoder masks at each stage. Comparative experiments were conducted between the proposed MSGU-Net and state-of-the-art networks on the ISIC2017 and ISIC2018 datasets. In short, compared to the baseline U-Net, our model achieves superior segmentation performance while reducing parameter and computation costs by 96.08 and 92.59%, respectively. Moreover, MSGU-Net can serve as a lightweight deep neural network suitable for deployment across a range of intelligent devices and mobile platforms, offering considerable potential for widespread adoption. [ABSTRACT FROM AUTHOR]

In order to address issues such as high noise, low brightness, and blurred details in low-light conditions, a new algorithm named UMDCEAD-NET, integrating zero-reference depth curves for low-light image enhancement and denoising, was developed. The algorithm's design was initially centered around a feature extraction network, employing a U-Net architecture as the backbone network. To enhance the feature extraction capabilities and preserve more detailed image information, Mobile-Net was integrated into the downsampling phase of the U-Net backbone. Subsequently, to address the issue of inadequate lighting and pixellevel image degradation, the extracted features underwent iteration using depth curve estimation (LE-curve), in conjunction with depth separable convolution, which served to reduce the network model's parameter count. Furthermore, five non-reference loss functions were engineered to bolster the algorithm's generalization capabilities and its retention of detail under varying lighting conditions. The enhanced image was then subjected to noise reduction in tandem with AD-NET(attentional denoising network), thereby diminishing the noise and aligning the image more closely with human visual perception. Experimental outcomes demonstrated that the proposed algorithm achieved an average PSNR (peak signal-to-noise ratio) of 22. 29 on the public dataset Zero-DCE, which exceeded the performance of the Zero-DCE ++ algorithm by 32%. Additionally, on the public dataset LOL, the algorithm attained an average PSNR of 21.15, outperforming the SGZ algorithm by 3%. These results indicate that the algorithm effectively mitigates noise in enhanced images, enriching the detail information in both dark and bright regions, and significantly improving image quality compared to other mainstream algorithms. [ABSTRACT FROM AUTHOR]

The transient electromagnetic (TEM) method is a crucial tool for subsurface exploration, providing essential insights into the electrical resistivity structures beneath the Earth's surface. Traditional forward modeling approaches, such as the finite-difference time-domain (FDTD) method and the finite-element method (FEM), are computationally intensive, limiting their practicality for real-time, high-resolution, or large-scale investigations. To address these challenges, we present Deep-TEMNet, an advanced deep learning framework specifically designed for two-dimensional TEM forward modeling. Deep-TEMNet integrates the U-Net architecture with a tailored two-dimensional long short-term memory (2D LSTM) module, allowing it to effectively capture complex spatial-temporal relationships in TEM data. The U-Net component enables high-resolution spatial feature extraction, while the 2D LSTM module enhances temporal modeling by processing spatial sequences in two dimensions, thereby optimizing the representation of electromagnetic field dynamics over time. Trained on high-fidelity FEM-generated datasets, Deep-TEMNet achieves exceptional accuracy in reproducing electromagnetic field distributions across diverse geological scenarios, with a mean squared error of 0.00000134 and a root mean square percentage error of 0.002373019. The framework offers over 150 times the computational speed of traditional FEMs, with an average inference time of just 3.26 s. Extensive validation across varied geological conditions highlights Deep-TEMNet's robustness and adaptability, establishing its potential for efficient, large-scale subsurface mapping and real-time data processing. By combining U-Net's spatial resolution capabilities with the sequential processing strength of the 2D LSTM module, Deep-TEMNet significantly advances computational efficiency and accuracy, positioning it as a valuable tool for geophysical exploration, environmental monitoring, and other applications requiring scalable, real-time TEM analyses that are easily integrated into remote sensing workflows. [ABSTRACT FROM AUTHOR]

Coastal wetlands are critical for global biodiversity and ecological stability, yet the invasive Spartina alterniflora (S. alterniflora) poses severe threats to these ecosystems. This study evaluates the effectiveness of management efforts targeting S. alterniflora in the Yellow River Delta (YRD) using Zhuhai-1 hyperspectral imagery and the U-Net method. The U-Net model, coupled with the Relief-F algorithm, achieved a superior extraction accuracy (Kappa > 0.9 and overall accuracy of 93%) compared to traditional machine learning methods. From 2019 to 2021, S. alterniflora expanded rapidly, increasing from 4055.06 hm2 to 6105.50 hm2, primarily in tidal flats and water bodies. A clearing project reduced its extent to 5063.62 hm2 by 2022, and by 2023, only 0.55 hm2 remained. These results underscore the effectiveness of Shandong's management policies but highlight the risk of regrowth due to the species' resilience. Continuous monitoring and maintenance are essential to prevent its resurgence and ensure wetland restoration. This study offers critical insights into dynamic vegetation monitoring and informs conservation strategies for wetland health. [ABSTRACT FROM AUTHOR]

The task of nucleus segmentation plays an important role in medical image analysis. However, due to the challenge of detecting small targets and complex boundaries in datasets, traditional methods often fail to achieve satisfactory results. Therefore, a novel nucleus segmentation method based on the U-Net architecture is proposed to overcome this issue. Firstly, we introduce a Weighted Feature Enhancement Unit (WFEU) in the encoder decoder fusion stage of U-Net. By assigning learnable weights to different feature maps, the network can adaptively enhance key features and suppress irrelevant or secondary features, thus maintaining high-precision segmentation performance in complex backgrounds. In addition, to further improve the performance of the network under different resolution features, we designed a Double-Stage Channel Optimization Module (DSCOM) in the first two layers of the model. This DSCOM effectively preserves high-resolution information and improves the segmentation accuracy of small targets and boundary regions through multi-level convolution operations and channel optimization. Finally, we proposed an Adaptive Fusion Loss Module (AFLM) that effectively balances different lossy targets by dynamically adjusting weights, thereby further improving the model's performance in segmentation region consistency and boundary accuracy while maintaining classification accuracy. The experimental results on 2018 Data Science Bowl demonstrate that, compared to state-of-the-art segmentation models, our method shows significant advantages in multiple key metrics. Specifically, our model achieved an IOU score of 0.8660 and a Dice score of 0.9216, with a model parameter size of only 7.81 M. These results illustrate that the method proposed in this paper not only excels in the segmentation of complex shapes and small targets but also significantly enhances overall performance at lower computational costs. This research offers new insights and references for model design in future medical image segmentation tasks. [ABSTRACT FROM AUTHOR]

To solve the problem of low segmentation model accuracy due to the complex shape of carbon slag in the aluminum electrolysis fire-eye image and the blurring of the boundary between the slag and the surrounding electrolyte, this paper proposes a segmentation model of the fire-eye image based on an improved U-Net. The model reduces the depth of the traditional U-Net to four layers and uses the multiscale dilated convolution module (MDCM) in the down-sampling stage. Second, the Convolutional Block Attention Module (CBAM) is embedded in the skip connection part of the network to improve the ability of the model to extract contextual features from images of multiple scales, enhance the guidance of high-level features to low-level features, and make the model pay more attention to the critical regions. To alleviate the negative impact of the imbalance of positive and negative examples in the dataset, the weighted binary cross-entropy loss and the Dice loss are used to replace the traditional cross-entropy loss. The experimental results show that the segmentation accuracy of the improved model on the fire-eye dataset reaches 88.03%, which is 5.61 percentage points higher than U-Net. [ABSTRACT FROM AUTHOR]

The three-dimensional (3D) reconstruction of the hepatic duct tree is significant for the minimally invasive surgery of hepatobiliary stone disease, which can be influenced by the quantity of the CT scans of hepatic ducts. If insufficient CT scans with low inter-slice resolution are directly utilized for 3D reconstruction, discontinuities and gaps will emerge in the reconstructed hepatic duct tree. In this paper, a novel end-to-end deep learning framework is designed for the inter-slice super-resolution segmentation of the CT slices of hepatic ducts, which can improve the 3D reconstruction performance in the inter-slice dimension. Specifically, the framework cascades into an inter-slice super-resolution subnetwork and a segmentation subnetwork. A deep learning network is introduced as the inter-slice super-resolution subnetwork to generate an intermediate slice between two adjacent CT slices in the simulated CT scans with low inter-slice resolution. To capture the spatiotemporal correlation existing in the CT scans of hepatic ducts, the ConvLSTM is introduced into the U-Net-like segmentation subnetwork in the high-dimensional feature space. To further suppress the problems of discontinuities and gaps, a structure-aware loss function is proposed by incorporating the structural similarity index measure (SSIM) as a regulator to dynamically assign the contribution of the generated CT slice to the total loss of the designed framework. Experimental results demonstrate that our proposed framework performs better segmentations for hepatic ducts than several existing deep learning networks with the performance of 0.7690 DICE and a 0.7712 F1-score, which is beneficial for the 3D reconstruction of the hepatic duct tree. [ABSTRACT FROM AUTHOR]

Image hiding aims to secretly embed secret information into a cover image and then recover the hidden data with minimal or no loss at the receiving end. Many works on steganography and deep learning have proved the huge prospects of deep learning in the field of image information hiding. However, current deep learning-based steganography research exposes significant limits, among which key issues such as how to improve embedding capacity, imperceptibility, and robustness remain crucial for image-hiding tasks. This article introduces PUA-Net, a new end-to-end neural network model for image steganography. PUA-Net consists of three main components: 1) the CbDw attention module, 2) the attention gate module, and 3) the partial combination convolution module. Each of these components utilizes structural reparameterization operations. In addition, we propose a residual image minimization loss function and use a combination of loss functions based on this loss function. This model can seamlessly embed bit stream information of different capacities into images to generate stego images that are imperceptible to the human eye. Experimental results confirm the effectiveness of our model, achieving an RS-BPP of 5.98 when decoding the extracted secret information and recovering the cover image. When only the extracted secret information is output, the model achieves a maximum RS-BPP of 6.94. Finally, experimental results show that our PUA-Net model outperforms deep learning-based steganography approaches on COCO, ImageNet, and BOSSbase datasets, including GAN-based methods such as Stegastamp and SteganoGAN. [ABSTRACT FROM AUTHOR]

Automatic segmentation of various hepatic structures from computed tomography images is of great importance in surgery and treatment planning for patients diagnosed with liver cancer. This work tends to identify the optimal semantic segmentation architecture for the task of segmentation of hepatic structures from computed tomography (CT) images. Segmentation of hepatic structures from CT images is carried out using four popular semantic segmentation architectures viz. fully convolutional network (FCN), SegNet, U-Net and DeepLabV3 + with various encoder configurations on 3D-ircadb-01, CHAOS and on an institutional dataset images. Segmentation of liver, liver tumour, hepatic artery, portal vein and hepatic vein are carried out simultaneously using a single network as a dense pixel prediction task. An extensive experimental assessment of hepatic structures segmentation was carried out using 13 networks derived from the aforesaid 4 segmentation architectures. The segmentation performance is assessed in terms of segmentation accuracy, dice coefficient, Jaccard index and boundary F1score. Experimental results show that DeepLabV3 + architecture with Xception encoder records better segmentation performance with dice coefficient of 98.13% 98.87% and 98.92% on 3D-ircadb-01, CHAOS and on the institutional datasets respectively. The comparison with the state-of-the-art techniques presented exemplifies the effectiveness of the fully depth separable convolutional network on the segmentation of various hepatic structures. Depth separable convolution applied to atrous spatial pyramid pooling (ASPP) with multiple sampling rates at the decoder of the DeepLabV3 + architecture together with the entirely separable convolution based Xception model as encoder stands as evidence towards the exceptional segmentation performance of hepatic structures. [ABSTRACT FROM AUTHOR]

Semantic segmentation of high-resolution images from remote sensing is crucial across various sectors. However, due to limitations in computational resources and the complexity of network architectures, many sophisticated semantic segmentation models struggle with efficiency in real-world applications, leading to an interest in developing lightweight model like borders. These models often employ a dual-branch structure, which balances processing speed and performance effectively. Yet, this design typically falls short in leveraging shallow structural information to enrich the dual branches with comprehensive multiscale data. Additionally, the lightweight components struggle to capture the global contextual details of feature sets efficiently. When compared to state-of-the-art models, lightweight semantic segmentation models usually exhibit performance gaps. To address these issues, we introduce a novel approach that incorporates a deep-shallow interaction mechanism with an attention module to improve water body segmentation efficiency. This method spatially adjusts feature representations to better identify water-related data, utilizing a U-Net frame work to enhance the accuracy of edge detection in water zones by providing more precise local positioning information. The attention mechanism processes and merges low and high-level data separately in different dimensions, allowing for the effective distinction of water areas from their surroundings by blending spatial attributes with in-depth context insights. Experimental outcomes demonstrate a remarkable 95% accuracy, showcasing the proposed method's superiority over existing models. [ABSTRACT FROM AUTHOR]

BACKGROUND: Liver cancer poses a significant health challenge due to its high incidence rates and complexities in detection and treatment. Accurate segmentation of liver tumors using medical imaging plays a crucial role in early diagnosis and treatment planning. OBJECTIVE: This study proposes a novel approach combining U-Net and ResNet architectures with the Adam optimizer and sigmoid activation function. The method leverages ResNet's deep residual learning to address training issues in deep neural networks. At the same time, U-Net's structure facilitates capturing local and global contextual information essential for precise tumor characterization. The model aims to enhance segmentation accuracy by effectively capturing intricate tumor features and contextual details by integrating these architectures. The Adam optimizer expedites model convergence by dynamically adjusting the learning rate based on gradient statistics during training. METHODS: To validate the effectiveness of the proposed approach, segmentation experiments are conducted on a diverse dataset comprising 130 CT scans of liver cancers. Furthermore, a state-of-the-art fusion strategy is introduced, combining the robust feature learning capabilities of the UNet-ResNet classifier with Snake-based Level Set Segmentation. RESULTS: Experimental results demonstrate impressive performance metrics, including an accuracy of 0.98 and a minimal loss of 0.10, underscoring the efficacy of the proposed methodology in liver cancer segmentation. CONCLUSION: This fusion approach effectively delineates complex and diffuse tumor shapes, significantly reducing errors. [ABSTRACT FROM AUTHOR]

Simple Summary: In vivo dose monitoring during treatment is crucial for successfully implementing boron neutron capture therapy in clinical practice but still remains not feasible. This study investigates the potentialities of Compton imaging in this setting by employing Monte Carlo simulation techniques, considering the application of deep convolutional neural network architectures to reduce image degradations in few-iteration reconstructed images obtained with the standard maximum-likelihood expectation-maximization algorithm, pursuing the avoidance of the iteration time associated with many-iterations reconstructions, enabling a prompt dose reconstruction during the treatment. The research examines the use of the standard U-Net architecture and two variants based on the deep convolutional framelets framework to accomplish this task, showing promising results in terms of reconstruction accuracy and processing time. Background: Boron neutron capture therapy (BNCT) is an innovative binary form of radiation therapy with high selectivity towards cancer tissue based on the neutron capture reaction 10B(n, α)7Li, consisting in the exposition of patients to neutron beams after administration of a boron compound with preferential accumulation in cancer cells. The high linear energy transfer products of the ensuing reaction deposit their energy at the cell level, sparing normal tissue. Although progress in accelerator-based BNCT has led to renewed interest in this cancer treatment modality, in vivo dose monitoring during treatment still remains not feasible and several approaches are under investigation. While Compton imaging presents various advantages over other imaging methods, it typically requires long reconstruction times, comparable with BNCT treatment duration. Methods: This study aims to develop deep neural network models to estimate the dose distribution by using a simulated dataset of BNCT Compton camera images. The models pursue the avoidance of the iteration time associated with the maximum-likelihood expectation-maximization algorithm (MLEM), enabling a prompt dose reconstruction during the treatment. The U-Net architecture and two variants based on the deep convolutional framelets framework have been used for noise and artifact reduction in few-iteration reconstructed images. Results: This approach has led to promising results in terms of reconstruction accuracy and processing time, with a reduction by a factor of about 6 with respect to classical iterative algorithms. Conclusions: This can be considered a good reconstruction time performance, considering typical BNCT treatment times. Further enhancements may be achieved by optimizing the reconstruction of input images with different deep learning techniques. [ABSTRACT FROM AUTHOR]

A plain-architecture and effective image dehazing scheme, called Encoder-Decoder Network with Guided Transmission Map (EDN-GTM), is proposed in this paper. Nowadays, neural networks are often built based on complex architectures and modules, which inherently prevent them from being efficiently deployed on general mobile platforms that are not integrated with latest deep learning operators. Hence, from a practical point of view, plain-architecture networks would be more appropriate for implementation. To this end, we aim to develop non-sophisticated networks with effective dehazing performance. A vanilla U-Net is adopted as a starting baseline, then extensive analyses have been conducted to derive appropriate training settings and architectural features that can optimize dehazing effectiveness. As a result, several modifications are applied to the baseline such as plugging spatial pyramid pooling to the bottleneck and replacing ReLU activation with Swish activation. Moreover, we found that the transmission feature estimated by Dark Channel Prior (DCP) can be utilized as an additional prior for a generative network to recover appealing haze-free images. Experimental results on various benchmark datasets have shown that the proposed EDN-GTM scheme can achieve state-of-the-art dehazing results as compared to prevailing dehazing methods which are built upon complex architectures. In addition, the proposed EDN-GTM model can be combined with YOLOv4 to witness an improvement in object detection performance in hazy weather conditions. The code of this work is publicly available at https://github.com/tranleanh/edn-gtm. [ABSTRACT FROM AUTHOR]

Multi-stage networks function by applying the concept of cascading, which alleviates the difficulties of network structure optimization using the single-stage method. Image deblurring methods based on multi-stage networks have previously been proposed and provided satisfactory results. However, owing to an excessive reliance on stacked subnets and residual blocks, most existing image deblurring methods that employ multi-stage networks suffer from two drawbacks: a complicated network structure and insufficient image representation of the model. To avoid these constraints, a novel multi-stage network deblurring model is proposed. Lightweight subnets are embedded in each stage of the model to gradually learn input image characteristics, which facilitates process optimization. The Ghost module is introduced as the basic unit of a neural network to reduce the required number of calculations and parameters. A wavelet reconstruction module is also added to avoid loss of image details. Finally, a more comprehensive loss function is designed to improve the quality of the generated images. Experimental results obtained using the GoPro dataset show that the proposed deblurring model achieves satisfactory performance in terms of both subjective results and objective evaluation. [ABSTRACT FROM AUTHOR]

Due to the limited availability of high-quality optical images during the rice growth period in the Lingnan region of China, effectively monitoring the rice planting situation has been a challenge. In this study, we utilized multi-temporal Sentinel-1 data to develop a method for rapidly extracting the range of rice fields using a threshold segmentation approach and employed a U-Net deep learning model to delineate the distribution of rice fields. Spatio-temporal changes in rice distribution in Leizhou City, Guangdong Province, China, from 2017 to 2021 were analyzed. The results revealed that by analyzing SAR-intensive time series data, we were able to determine the backscattering coefficient of typical crops in Leizhou and use the threshold segmentation method to identify rice labels in SAR-intensive time series images. Furthermore, we extracted the distribution range of early and late rice in Leizhou City from 2017 to 2021 using a U-Net model with a minimum relative error accuracy of 3.56%. Our analysis indicated an increasing trend in both overall rice planting area and early rice planting area, accounting for 44.74% of early rice and over 50% of late rice planting area in 2021. Double-cropping rice cultivation was predominantly concentrated in the Nandu River basin, while single-cropping areas were primarily distributed along rivers and irrigation facilities. Examination of the traditional double-cropping areas in Fucheng Town from 2017 to 2021 demonstrated that over 86.94% had at least one instance of double cropping while more than 74% had at least four instances, which suggested that there is high continuity and stability within the pattern of rice cultivation practices observed throughout Leizhou City. [ABSTRACT FROM AUTHOR]

Deformable image registration minimizes the discrepancy between moving and fixed images by establishing linear and nonlinear spatial correspondences. It plays a crucial role in surgical navigation, image fusion and disease analysis. Its challenge lies in the large number of deformed parameters and the uncertainty of acquisition conditions. Benefiting from the powerful ability to capture hierarchical features and spatial relationships of convolutional neural networks, the medical image registration task has made great progress. Nowadays, the long-range relationship modeling and adaptive selection of self-attention show great potential and have also attracted much attention from researchers. Inspired by this, we propose a new method called Multi-scale Large Kernel Attention UNet (MLKA-Net), which combines a large kernel convolution with the attention mechanism using a multi-scale strategy, and uses a correction module to fine-tune the deformation field to achieve high-accuracy registration. Specifically, we first propose a multi-scale large kernel attention mechanism (MLKA), which generates attention maps by aggregating information from convolution kernels at different scales to improve local feature modeling capabilities of attention. Furthermore, we employ large kernel dilation convolution in proposed attention to construct sufficiently long-range relationships, while keeping lower number of parameters. Finally, to further improve local accuracy of the registration, we design an additional correction module and unsupervised framework to fine-tune the deformation field to solve the issue of original information loss in multilayer networks. Our method is compared qualitatively and quantitatively with 24 representative and advanced methods on the 3 public available 3D datasets from IXI database, LPBA40 dataset and OASIS database, respectively. The experiments demonstrate the excellent performance of the proposed method. [ABSTRACT FROM AUTHOR]

The U-Net architecture, which performs segmentation on images, has also achieved very successful results in the medical field. However, there is also a need to improve the U-Net architecture for better results. In this article, some improvement proposals are presented for the U-Net model's encoder part, and the segmentation success of the implemented architecture for segmentation of dermoscopic image lesions is evaluated. The PH2 dataset and the "International Skin Imaging Collaboration" datasets (ISIC-2016 and ISIC-2017) were used for the research. The traditional data augmentation method was applied to the selected PH2 dataset samples. The results of the proposed model (EnecaU-Net) and the U-Net model obtained with the PH2 dataset were compared. Furthermore, in this article, the mix data augmentation method, which has an influence on the model's segmentation success, is examined for lesion segmentation on dermoscopic images. This investigation was made with the ISIC-2016 dataset, and its experimental results were compared with the same amount of the ISIC-2017 dataset that didn't apply data augmentation operations. Although, during the evaluation phase, Dice and Jaccard (IoU) metrics were used primarily to measure the success of the model, specificity, sensitivity, and accuracy criteria were also used. According to our results, the segmentation success of the EnecaU-Net model applied for lesion segmentation on dermoscopic images is high, and the applied mix data augmentation method improves the segmentation success of the EnecaU-Net model. The average test results achieved by the proposed model are 88.05% and 80.30% for ISIC2016 and 83.09% and 74.54% for ISIC-2017 in terms of the Dice and Jaccard values, respectively. [ABSTRACT FROM AUTHOR]

Developmental hip dysplasia (DDH) is a disease in which the hip joint fails to develop normally due to various causes before, during or after birth. The most important method used for the detection of DDH is hip ultrasonography. The stage of obtaining the hip US image varies because it depends on the operator and external influences. In this study, an artificial intelligence-based system has been developed to eliminate this variability. The developed system includes a 2-stage deep learning model. The main purpose of the system is to automatically determine whether the US images obtained by physicians are suitable for the calculation of alpha and beta angles required for diagnosis. The system uses the U-NET architecture in the first stage and the masked region-based convolutional neural network (MBT-ESA) architecture in the second stage. For the training, 540 images were taken from Selçuk University Faculty of Medicine hospital with the approval of the ethics committee. A total of 840 images were obtained for training with data augmentation. U-NET architecture training resulted in an accuracy of 0.93 and region-based convolutional neural network training with mask resulted in an accuracy of 0.96. The overall system accuracy was calculated as 0.96. The results obtained in this study suggest that by increasing the number of real-time tests and images, the inter-operator variability in the diagnosis of DDH can be eliminated. [ABSTRACT FROM AUTHOR]

Developmental toxicity (DevTox) tests evaluate the adverse effects of chemical exposures on an organism's development. Although current testing primarily relies on large mammalian models, the emergence of new approach methodologies (NAMs) is encouraging industries and regulatory agencies to evaluate novel assays. C. elegans have emerged as NAMs for rapid toxicity testing because of its biological relevance and suitability to high throughput studies. However, current low-resolution and labor-intensive methodologies prohibit its application for sub-lethal DevTox studies at high throughputs. With the recent advent of the large-scale microfluidic device, vivoChip, we can now rapidly collect 3D high-resolution images of ~ 1000 C. elegans from 24 different populations. While data collection is rapid, analyzing thousands of images remains time-consuming. To address this challenge, we developed a machine-learning (ML)-based image analysis platform using a 2.5D U-Net architecture (vivoBodySeg) that accurately segments C. elegans in images obtained from vivoChip devices, achieving a Dice score of 97.80%. vivoBodySeg processes 36 GB data per device, phenotyping multiple body parameters within 35 min on a desktop PC. This analysis is ~ 140 × faster than the manual analysis. This ML approach delivers highly reproducible DevTox parameters (4–8% CV) to assess the toxicity of chemicals with high statistical power. [ABSTRACT FROM AUTHOR]

Image dehazing is an ill-posed issue in low-level computer vision; therefore, it grabbed many researchers' attention. The key mechanism to improve dehazing performance remains unclear, although many existing network pipelines work fine. To improve the performance of the image dehazing network, a hierarchical model named "Selective Attentive Wavelet M-Net+" (SWAM-Net+) was proposed. In order to enrich the features from the wavelet domain, a "Selective Wavelet Attentive Module" was introduced in M-Net+. Several key components of our network are used for extracting the multiscale features through parallel multi-resolution convolution channels. Contextual information is collected using a dual attention unit, and the attention is based on multiscale feature aggregation. We replaced summation and concatenation operations by introducing the Selective Kernel Feature Fusing module to achieve feature aggregation. Furthermore, our network achieves comprehensively better performance results on the RESIDE dataset both qualitatively and quantitatively. [ABSTRACT FROM AUTHOR]

Roads are an essential element of various information systems such as geographic information systems, transportation systems, etc. The main source of road information is remote sensing data as it covers a large amount of area. Despite recent technological advancements precise road information extraction is still a tedious task. In this work, a computational-efficient deep learning architecture AM-Unet is proposed to extract road information from high-resolution aerial imagery. The proposed method alters the design of Unet architecture for the encoder, decoder, and skip connections. These changes enhance the computational efficiency of the decoder to recapture spatial location information. The experiments are performed on complex high-resolution (HR) aerial images and the results are assessed on diverse quantitative parameters. The experimental results are compared to other deep learning methods which reflects the improvement in results on Precision, recall, Acc and F1-score parameters. [ABSTRACT FROM AUTHOR]

Applying deep learning (DL) for annotating surgical instruments in robot-assisted minimally invasive surgeries (MIS) represents a significant advancement in surgical technology. This systematic review examines 48 studies that utilize advanced DL methods and architectures. These sophisticated DL models have shown notable improvements in the precision and efficiency of detecting and segmenting surgical tools. The enhanced capabilities of these models support various clinical applications, including real-time intraoperative guidance, comprehensive postoperative evaluations, and objective assessments of surgical skills. By accurately identifying and segmenting surgical instruments in video data, DL models provide detailed feedback to surgeons, thereby improving surgical outcomes and reducing complication risks. Furthermore, the application of DL in surgical education is transformative. The review underscores the significant impact of DL on improving the accuracy of skill assessments and the overall quality of surgical training programs. However, implementing DL in surgical tool detection and segmentation faces challenges, such as the need for large, accurately annotated datasets to train these models effectively. The manual annotation process is labor-intensive and time-consuming, posing a significant bottleneck. Future research should focus on automating the detection and segmentation process and enhancing the robustness of DL models against environmental variations. Expanding the application of DL models across various surgical specialties will be essential to fully realize this technology's potential. Integrating DL with other emerging technologies, such as augmented reality (AR), also offers promising opportunities to further enhance the precision and efficacy of surgical procedures. [ABSTRACT FROM AUTHOR]

Exploration and production activities in deep-water oil and gas reservoirs can directly impact the surrounding ecosystems. Thus, a tool capable of measuring oil and gas leaks based on surveillance images, especially in pre-mature stages, is of great importance for ensuring safety and environmental protection. In the present work, a Convolutional Neural Network (U-Net) is applied to leak images using transfer learning and hyperparameter optimization, aiming to predict bubble diameter and flow rate. The data were extracted from a reduced model leak experiment, with a total of 77,676 frames processed, indicating a Big Data context. The results agreed with the data obtained in the laboratory: for the flow rate prediction, coefficients of determination by transfer learning and hyperparameter optimization were, respectively, 0.938 and 0.941. Therefore, this novel methodology has potential applications in the oil and gas industry, in which leaks captured by a camera are measured, supporting decision-making in the early stages and building a framework of a mitigation strategy in industrial environments. [ABSTRACT FROM AUTHOR]

Video saliency prediction aims to simulate human visual attention by locating the most pertinent and instructive areas within a video frame or sequence. While ignoring the audio aspect, time and space data are essential when measuring video saliency, especially with challenging factors like swift motion, changeable background, and nonrigid deformation. Additionally, video saliency detection is inappropriate when using image saliency models directly neglecting video temporal information. This paper suggests a novel Bidirectional Multi-scale SpatioTemporal Network (BMST-Net) for identifying prominent video objects to address the above problem. The BMST-Net yields notable results for any given frame sequence, employing an encoder and decoder technique to learn and map features over time and space. The BMST-Net model consists of bidirectional LSTM (Long Short-Term Memory) and CNN (Convolutional Neural Network), where the VGG16 (Visual Geometry Group) single layer is used for feature extraction of the input video frames. Our proposed approach produced noteworthy findings concerning qualitative and quantitative investigation of the publicly available challenging video datasets, achieving competitive performance concerning state-of-the-art saliency models. [ABSTRACT FROM AUTHOR]

Cervical cancer is a significant health issue for women and ranks fourth in the world among the most dangerous cancers. An automatic diagnostic system is needed for pap smears to assist medical experts in diagnosing cervical cancer. One of the automatic diagnosis systems in detecting cervical cancer is semantic segmentation. Convolutional Neural Networks (CNN), particularly the U-Net architecture, have been widely used for segmentation tasks in medical imaging. Although U-Net has demonstrated effectiveness, its performance on low-quality images is often suboptimal, with issues such as loss of fine details during the down-sampling process. This study combines image enhancement and Double Dropout USeg-Net (DDUSeg-Net). Image enhancement techniques are applied to pap-smear images to improve image quality such as Gamma Correction for enhanced contrast, and Median Filtering for reduced noise. The proposed DDUSeg-Net architecture builds on the U-Net model by incorporating two U-Net blocks for more detailed feature extraction. SegNet's pooling indices are added to preserve spatial information during the segmentation process. Additionally, dropout layers are introduced to prevent overfitting and reduce the model's overall complexity. The image enhancement results indicate that the Mean Squared Error (MSE), Peak Signal to Ratio (PNSR), and Structural Similarity Image Index (SSIM) are above 85%. The performance metrics for the DDUSeg-Net model obtained accuracy, precision, recall, and F1-score above 90%. This analysis used 2D pap-smear images from the Herlev dataset. Overall, the combination of image enhancement and DDUSegNet demonstrates strong robustness in the segmentation of pap-smear images, effectively balancing the detection of the intersection areas between the nucleus, cytoplasm, and background. [ABSTRACT FROM AUTHOR]

Retinal image segmentation is crucial for the early diagnosis of some diseases like diabetes and hypertension. Current methods face many challenges, such as inadequate multi-scale semantics and insufficient global information. In view of this, we propose a network called multi-scale semantics mining and tiny details enhancement (MSM-TDE). First, a multi-scale feature input module is designed to capture multi-scale semantics information from the source. Then a fresh multi-scale attention guidance module is constructed to mine local multi-scale semantics while a global semantics enhancement module is proposed to extract global multi-scale semantics. Additionally, an auxiliary vessel detail enhancement branch using dynamic snake convolution is built to enhance the tiny vessel details. Extensive experimental results on four public datasets validate the superiority of MSM-TDE, which obtains competitive performance with satisfactory model complexity. Notably, this study provides an innovative idea of multi-scale semantics mining by diverse methods. [ABSTRACT FROM AUTHOR]

The examination of power transmission lines using UAVs (Unmanned Aerial Vehicles) is crucial for ensuring grid security. However, it is challenging for existing deep learning models to achieve a balance between accuracy and efficiency in recognizing power lines, especially when they are affected by intricate environmental backdrops and the thin structure of power lines. To address this issue, this paper proposes an improved model based on U-Net (PL-UNet), which aims to improve the ability of UAVs to recognize power lines in complex environments in real time. To reduce the model parameters, the lightweight EfficientNetV2-S is chosen as the encoder. To address the issues of information redundancy and insufficient local structure capture caused by the skip connections, we propose a multi-scale attention gate (MSAG) in the decoding part to improve the accuracy of key region feature extraction with less computational cost. Meanwhile, the dynamic weighted fusion (DWF) module is designed to effectively fuse the features through adaptive weighting to improve the flexibility of feature expression. After completing feature fusion, we further introduce a lightweight cross-stage partial pyramid block (CPPB) module, which performs multi-scale enhancement and channel optimization of the fused features through integrating multi-scale convolutional operations and separating and fusing feature channels. Finally, the hybrid loss function of weighted cross-entropy and dice is used to solve the category imbalance problem. Comparison experiments and ablation analysis are performed on top of the power line public dataset. The proposed PL-UNet has achieved 79.98% mIoU, with a parameter count of 21.27M and a detection speed of 56.86 fps. This shows that the network has a good real-time segmentation performance. [ABSTRACT FROM AUTHOR]

Background: Retinal blood vessel segmentation plays an important role in diagnosing retinal diseases such as diabetic retinopathy, glaucoma, and hypertensive retinopathy. Accurate segmentation of blood vessels in retinal images presents a challenging task due to noise, low contrast, and the complex morphology of blood vessel structures. Methods: In this study, we propose a novel ensemble learning framework combining four deep learning architectures: U-Net, ResNet50, U-Net with a ResNet50 backbone, and U-Net with a transformer block. Each architecture is customized to enhance feature extraction and segmentation performance. The models are trained on the DRIVE and STARE datasets to improve the degree of generalization and evaluated using the performance metric accuracy, F1-Score, sensitivity, specificity, and AUC. Results: The ensemble meta-model integrates predictions from these architectures using a stacking approach, achieving state-of-the-art performance with an accuracy of 0.9778, an AUC of 0.9912, and an F1-Score of 0.8231. These results demonstrate the performance of the proposed technique in identifying thin retinal blood vessels. Conclusions: A comparative analysis using qualitative and quantitative results with individual models highlights the robustness of the ensemble framework, especially under conditions of noise and poor visibility. [ABSTRACT FROM AUTHOR]

Purpose: To shorten CEST acquisition time by leveraging Z‐spectrum undersampling combined with deep learning for CEST map construction from undersampled Z‐spectra. Methods: Fisher information gain analysis identified optimal frequency offsets (termed "Fisher offsets") for the multi‐pool fitting model, maximizing information gain for the amplitude and the FWHM parameters. These offsets guided initial subsampling levels. A U‐NET, trained on undersampled brain CEST images from 18 volunteers, produced CEST maps at 3 T with varied undersampling levels. Feasibility was first tested using retrospective undersampling at three levels, followed by prospective in vivo undersampling (15 of 53 offsets), reducing scan time significantly. Additionally, glioblastoma grade IV pathology was simulated to evaluate network performance in patient‐like cases. Results: Traditional multi‐pool models failed to quantify CEST maps from undersampled images (structural similarity index [SSIM] <0.2, peak SNR <20, Pearson r <0.1). Conversely, U‐NET fitting successfully addressed undersampled data challenges. The study suggests CEST scan time reduction is feasible by undersampling 15, 25, or 35 of 53 Z‐spectrum offsets. Prospective undersampling cut scan time by 3.5 times, with a maximum mean squared error of 4.4e–4, r = 0.82, and SSIM = 0.84, compared to the ground truth. The network also reliably predicted CEST values for simulated glioblastoma pathology. Conclusion: The U‐NET architecture effectively quantifies CEST maps from undersampled Z‐spectra at various undersampling levels. [ABSTRACT FROM AUTHOR]

Background and Objectives: The increasing prevalence of skin cancer highlights the urgency for early intervention, emphasizing the need for advanced diagnostic tools. Computer-assisted diagnosis (CAD) offers a promising avenue to streamline skin cancer screening and alleviate associated costs. Methods: This study endeavors to develop an automatic segmentation system employing deep neural networks, seamlessly integrating data manipulation into the learning process. Utilizing an encoder-decoder architecture rooted in U-Net and augmented by wavelet transform, our methodology facilitates the generation of high-resolution feature maps, thus bolstering the precision of the deep learning model. Results: Performance evaluation metrics including sensitivity, accuracy, dice coefficient, and Jaccard similarity confirm the superior efficacy of our model compared to conventional methodologies. The results showed a accuracy of %96.89 for skin lesions in PH2 Database and %95.8 accuracy for ISIC 2017 database findings, which offers promising results compared to the results of other studies. Additionally, this research shows significant improvements in three metrics: sensitivity, Dice, and Jaccard. For the PH database, the values are 96, 96.40, and 95.40, respectively. For the ISIC database, the values are 92.85, 96.32, and 95.24, respectively. Conclusion: In image processing and analysis, numerous solutions have emerged to aid dermatologists in their diagnostic endeavors The proposed algorithm was evaluated using two PH datasets, and the results were compared to recent studies. Impressively, the proposed algorithm demonstrated superior performance in terms of accuracy, sensitivity, Dice coefficient, and Jaccard Similarity scores when evaluated on the same database images compared to other methods. [ABSTRACT FROM AUTHOR]

Introduction: Orthogonal Frequency Division Multiplexing (OFDM) is widely recognized for its high efficiency in modulation techniques and has been extensively applied in underwater acoustic communication. However, the unique sparsity and noise interference characteristics of the underwater channel pose significant challenges to the performance of traditional channel estimation methods. Methods: To address these challenges, we propose a sparse underwater channel estimation method that combines the Noise2Noise (N2N) algorithm with the Sparsity Adaptive Matching Pursuit (SAMP) algorithm. This novel approach integrates the N2N technique from image denoising theory with the SAMP algorithm, utilizing a constant iteration termination threshold that does not require prior information. The method leverages the U-net neural network structure to denoise noisy pilot signals, thereby restoring channel sparsity and enhancing the accuracy of channel estimation. Results: Simulation results indicate that our proposed method demonstrates commendable channel estimation performance across various signal-to-noise ratio (SNR) conditions. Notably, in low SNR environments, the N2N-SAMP algorithm significantly outperforms the traditional SAMP algorithm in terms of Mean Squared Error (MSE) and Bit Error Rate (BER). Specifically, at SNR levels of 0 dB, 10 dB, and 20 dB, the MSE of channel estimation is reduced by 58.95%, 76.08%, and 19.42%, respectively, compared to the SAMP algorithm that selects the optimal threshold based on noise power. Furthermore, the system's BER is decreased by 12.35%, 26.41%, and 29.62%, respectively. Discussion: The findings suggest that the integration of N2N and SAMP algorithms offers a promising solution for improving channel estimation in underwater communication channels, especially under low SNR conditions. The significant reduction in MSE and BER highlights the effectiveness of our proposed method in enhancing the reliability and accuracy of underwater communication systems. [ABSTRACT FROM AUTHOR]

Surface electromyography (sEMG) data has been extensively utilized in deep learning algorithms for hand movement classification. This paper aims to introduce a novel method for hand gesture classification using sEMG data, addressing accuracy challenges seen in previous studies. We propose a U-Net architecture incorporating a MobileNetV2 encoder, enhanced by a novel Bidirectional Long Short-Term Memory (BiLSTM) and metaheuristic optimization for spatial feature extraction in hand gesture and motion recognition. Bayesian optimization is employed as the metaheuristic approach to optimize the BiLSTM model's architecture. To address the non-stationarity of sEMG signals, we employ a windowing strategy for signal augmentation within deep learning architectures. The MobileNetV2 encoder and U-Net architecture extract relevant features from sEMG spectrogram images. Edge computing integration is leveraged to further enhance innovation by enabling real-time processing and decision-making closer to the data source. Six standard databases were utilized, achieving an average accuracy of 90.23% with our proposed model, showcasing a 3–4% average accuracy improvement and a 10% variance reduction. Notably, Mendeley Data, BioPatRec DB3, and BioPatRec DB1 surpassed advanced models in their respective domains with classification accuracies of 88.71%, 90.2%, and 88.6%, respectively. Experimental results underscore the significant enhancement in generalizability and gesture recognition robustness. This approach offers a fresh perspective on prosthetic management and human-machine interaction, emphasizing its efficacy in improving accuracy and reducing variance for enhanced prosthetic control and interaction with machines through edge computing integration. [ABSTRACT FROM AUTHOR]

The most common type of malignant brain tumor, gliomas, has a variety of grades that significantly impact a patient's chance of survival. Accurate segmentation of brain tumor regions from MRI images is crucial for enhancing diagnostic precision and refining surgical strategies. This task is particularly challenging due to the diverse sizes and shapes of tumors, as well as the intricate nature of MRI data. Mastering this segmentation process is essential for improving clinical outcomes and ensuring optimal treatment planning. In this research, we provide a UNet-based model (RR-UNet) designed specifically for brain tumor segmentation, which uses small and diverse datasets containing human-annotated ground truth segmentations. This model uses residual learning to improve segmentation results over the original UNet architecture, as shown by higher dice similarity coefficient (DSC) and Intersection over Union (IoU) scores. Residual blocks enable a deeper network, which can capture complex patterns. Residual blocks reuse features, allowing the network to learn more abstract and informative representations from input images. Through comprehensive evaluation and validation, we illustrate our method's efficacy and generalization capabilities, emphasizing its potential for real-world clinical applications. This segmentation model predicts DSC of 98.18% and accuracy of 99.78% in tumor segmentation using Figshare LGG (Low-grade glioma) FLAIR segmentation dataset and DSC of 98.54% & accuracy of 99.81% using BraTS 2020 dataset. The ablation study shows the importance of the model's residual mechanism. Overall, the proposed approach outperforms or compares to existing most recent algorithms in brain tumor segmentation tasks. [ABSTRACT FROM AUTHOR]

The major challenge of edge detection in denoised images is smoothing the edges, thus confusing algorithms to classify a true edge with false one. Existing learning and optimization methods tends to handle the above problem with selectivity approach to avoid the smoothed edges. In this paper, an efficient model to detect an edges of fuzzy noise images is introduced. Inspired by U-Net and dense network, a feature map reuse and residual learning with Dense Residual Network structure (DRNet) is proposed. The feature maps from earlier blocks are dynamically combined into succeeding layers as input for the proposed DRNet. This process not only aids in spatial reconstruction, but also enhances learning efficiency by utilizing more consistent gradients to allocate the most appropriate true edge. Furthermore, the noised pixel will be eliminated with the training process for the targeted edges. To overcome the overfitting challenge in the network, this paper proposes an implementation of the DropBlock technique. The model is trained and tested using a renowned edge detection public datasets. The experimental results demonstrate that the proposed model outperformed other conventional and learning detection algorithms in terms of peak signal-to-noise ratio in different levels. These results confirm the efficacy of this study and offer a fresh approach to enhancing edge detection. [ABSTRACT FROM AUTHOR]