Weighted Cross Entropy Research Articles

A domain adaptation framework for cross-modality SAR 3D reconstruction point clouds segmentation utilizing LiDAR data

Synthetic aperture radar (SAR) 3D point clouds reconstruction can eliminate the problems of layover in 2D SAR image projections, the recognition of the reconstructed point cloud can significantly enhance target identification and information extraction. Current SAR 3D point cloud segmentation methods, based on traditional machine learning clustering approaches, suffer from poor automation and accuracy. However, the absence of publicly labeled SAR datasets impedes the progress of deep learning-based SAR 3D point cloud segmentation methods. To tackle the aforementioned challenges, we introduce, for the first time, an alternative training approach for SAR 3D point cloud segmentation using LiDAR annotated data, which offers a more abundant sample pool, thus alleviating the lack of training sets. Nevertheless, a segmentation model trained on LiDAR point clouds exhibits a significant decline in performance when directly applied to SAR 3D reconstructed point clouds due to the cross-domain discrepancy. This research presents a pioneering domain adaptation 3D semantic segmentation framework to implement cross-modal learning for SAR point clouds. In our scheme, a simple yet effective technique is developed to achieve segmentation of SAR double-bounce scattering regions called SARDBS-Mix, which employs a mixing strategy, to capture the distinctive reflection characteristics of SAR data. Furthermore, we implement approaches including center alignment and normalization, local augmentation, and weighted cross entropy to mitigate LiDAR and SAR domain gap and class imbalances. The experimental results validate the feasibility and effectiveness of the proposed method for SAR 3D reconstructed point cloud segmentation.

A comprehensive theoretical framework for the optimization of neural networks classification performance with respect to weighted metrics

AbstractIn many contexts, customized and weighted classification scores are designed in order to evaluate the goodness of the predictions carried out by neural networks. However, there exists a discrepancy between the maximization of such scores and the minimization of the loss function in the training phase. In this paper, we provide a complete theoretical setting that formalizes weighted classification metrics and then allows the construction of losses that drive the model to optimize these metrics of interest. After a detailed theoretical analysis, we show that our framework includes as particular instances well-established approaches such as classical cost-sensitive learning, weighted cross entropy loss functions and value-weighted skill scores.

3DGraphSeg: A Unified Graph Representation- Based Point Cloud Segmentation Framework for Full-Range High-Speed Railway Environments

Point cloud semantic segmentation (PCSS) is crucial for digital twins of high-speed railways. By now, the concerned subjects are confined within the interior infrastructures of railways. However, the surrounding environments are also important for the safe operation. Concerning this issue, a full-range high-speed railway scanning scheme based on UAV borne LiDAR is utilized. However, the massive data volume and data distribution imbalance pose great challenges for PCSS. To address these issues, a novel PCSS framework called 3DGraphSeg is proposed in this paper. To cope with the massive data volume, a structural representation algorithm named Local Embedding Super-Point Graph (LE-SPG) is proposed to represent the vast point cloud into a concise graph while retain the data's inherent topology structure by local spatial embedding. Then, the Gated Integration Graph Convolutional Network (GIGCN) is proposed to contextual segment the graph. In GIGCN, to prevent the gradients from vanishing or exploding, the hidden states of Gated Recurrent Units (GRUs) in every layer are integrated using a new layer named Gated Hidden States Integration (GHSI). GHSI strengthens the back propagation by giving the loss function direct access to each layer and absorbs the features of different layers comprehensively, which enables the network to produce a smoother decision boundary and prevents the overfitting problem. Besides, to enhance its robustness to data imbalance, we propose a loss function: Adaptive Weighted Cross Entropy (AWCE). Finally, five experiments are designed for verification. The proposed framework has excelled in different datasets and outperforms state-of-the-art approaches on the SemanticRail dataset.

Synthetic MRI-Assisted and Self-Supervised Adaptive Segmentation of Organs-at-Risk (OARs) in MRI-Based Radiation Therapy

This study proposes a self-supervised solution for OAR segmentation, combining patch-based adaptation and unsupervised synthesis of T2-weighted MRI data to finetune the segmentation model. The aim is to improve adaptation to patient anatomy, overcome limited annotated MRI data, and enhance the generalizability of automatic segmentation models for gynecological cancers. The study used a patch-based cycle consistent generative adversarial network (cycle-GAN) for unsupervised MRI synthesis from CT scans of 20 patients, and a residual U-Net model for OARs segmentation. The segmentation model was trained and validated on synthetic MRI (sMRI) of 103 and 25 patient scans respectively, then finetuned on 78 MRI scans from radiation therapy fractions of 15 additional patients through three-fold cross validation. Self-supervised adaptation was applied, incorporating affine and elastic deformations, intensity shifting, and scaling. The model was trained on 96 × 96 × 96 sub-volumes and validated on entire pelvic sections of the same images. A combination of Dice and weighted cross entropy (CE) losses, with weights assigned for bladder (1), small bowel (1), rectum (2), sigmoid (2), left femoral head (0), and right femoral head (0), was used for OAR segmentation. The performance was evaluated against the model trained only on a limited number of acquired MRI data, as well as sMRI pretrained models with encoder weight freezing and either equal weighting or soft-tissue adjusted weighting. Our sMRI-assisted approach showed improved performance for challenging pelvic OARs compared to the method using only the acquired MRI data. The self-supervised fraction-adaptive segmentation results indicated better performance in target soft-tissues when using at least one treatment fraction for organ-specific adaptation. Our framework leverages pre-existing CT planning data for gynecological cancers to enhance the segmentation performance of OARs during MR-guided adaptive treatments. This approach offers substantial benefits for the radiation therapy workflow, including reduced variability in per-fraction segmentation and clinical burden. Further studies that involve human expert evaluations will be conducted to assess the practicality of this approach in radiation therapy.

Plane invariant segmentation of computed tomography images through weighted cross entropy optimized conditional GANs in compressed formats.

Computed tomography (CT) scan provides first-hand knowledge to doctors to identify an ailment. Deep neural networks help enhance image understanding through segmentation and labeling. In this work, we implement two variants of Pix2Pix generative adversarial networks (GANs) with varying complexities of generator and discriminator networks for plane invariant segmentation of CT scan images and subsequently propose an effective generative adversarial network with a suitably weighted binary cross-entropy loss function followed by image processing layer necessary for getting high-quality output segmentation. Our conditional GAN is powered by a unique set of an encoder-decoder network that coupled with the image processing layer produces enhanced segmentation. The network can be extended to the complete set of Hounsfield units and can also be implemented on smartphones. Furthermore, we also demonstrate effects on accuracy, F-1 score, and Jaccard index by using the conditional GAN networks on the spine vertebrae dataset, thus achieving an average of 86.28 % accuracy, 90.5 % Jaccard index score, and 89.9 % F-1 score in predicting segmented maps for validation input images. In addition, an overall lifting of accuracy, F-1 score, and Jaccard index graph for validation images with better continuity has also been highlighted.

Fast and Accurate Road Crack Detection Based on Adaptive Cost-Sensitive Loss Function.

Numerous detection problems in computer vision, including road crack detection, suffer from exceedingly foreground-background imbalance. Fortunately, modification of loss function appears to solve this puzzle once and for all. In this article, we propose a pixel-based adaptive weighted cross-entropy (WCE) loss in conjunction with Jaccard distance to facilitate high-quality pixel-level road crack detection. Our work profoundly demonstrates the influence of loss functions on detection outcomes and sheds light on the sophisticated consecutive improvements in the realm of crack detection. Specifically, to verify the effectiveness of the proposed loss, we conduct extensive experiments on four public databases, that is, CrackForest, AigleRN, Crack360, and BJN260. Compared to the vanilla WCE, the proposed loss significantly speeds up the training process while retaining the performance.

Weakly supervised perivascular spaces segmentation with salient guidance of Frangi filter.

To develop a weakly supervised 3D perivascular spaces (PVS) segmentation model that combines the filter-based image processing algorithm and the convolutional neural network. We present a weakly supervised learning method for PVS segmentation by combing a rule-based image processing approach Frangi filter with a canonical deep learning algorithm Unet using conditional random field theory. The weighted cross entropy loss function and the training patch selection were implemented for the optimization and to alleviate the class imbalance issue. The performance of the model was evaluated on the Human Connectome Project data. The proposed method increases the true positive rate compared to the rule-based method and reduces the false positive rate by 36% in the weakly supervised training experiment and 39.4% in the supervised training experiment compared to Unet, which results in superior overall performance. In addition, by training the model on manually quality controlled and annotated data which includes the subjects with the presence of white matter hyperintensities, the proposed method differentiates between PVS and white matter hyperintensities, which reduces the false positive rate by 78.5% compared to weakly supervised trained model. Combing the filter-based image processing algorithm and the convolutional neural network algorithm could improve the model's segmentation accuracy, while reducing the training dependence on the large scale annotated PVS mask data by the trained physician. Compared to the filter-based image processing algorithm, the data driven PVS segmentation model using quality-controlled data as the training target could differentiate the white matter hyperintensity from PVS resulting low false positive rate.

Region-wise loss for biomedical image segmentation

We propose Region-wise (RW) loss for biomedical image segmentation. Region-wise loss is versatile, can simultaneously account for class imbalance and pixel importance, and it can be easily implemented as the pixel-wise multiplication between the softmax output and a RW map. We show that, under the proposed RW loss framework, certain loss functions, such as Active Contour and Boundary loss, can be reformulated similarly with appropriate RW maps, thus revealing their underlying similarities and a new perspective to understand these loss functions. We investigate the observed optimization instability caused by certain RW maps, such as Boundary loss distance maps, and we introduce a mathematically-grounded principle to avoid such instability. This principle provides excellent adaptability to any dataset and practically ensures convergence without extra regularization terms or optimization tricks. Following this principle, we propose a simple version of boundary distance maps called rectified Region-wise (RRW) maps that, as we demonstrate in our experiments, achieve state-of-the-art performance with similar or better Dice coefficients and Hausdorff distances than Dice, Focal, weighted Cross entropy, and Boundary losses in three distinct segmentation tasks. We quantify the optimization instability provided by Boundary loss distance maps, and we empirically show that our RRW maps are stable to optimize. The code to run all our experiments is publicly available at: https://github.com/jmlipman/RegionWiseLoss.

A study on loss function against data imbalance in deep learning correction of precipitation forecasts

Sample imbalance has always been an important issue to be solved in heavy precipitation forecasts. Introducing appropriate constraints into the loss function of a machine learning model can mitigate the negative impact of sample imbalance on model training. This study explores the role of Dice loss function in dealing with sample imbalance and verifies its application on the correction of heavy precipitation forecasts using the U-Net neural network. On this basis, the application of ordinal classification in forecasts correction is further verified. The results show that the concept of Dice loss is highly similar to that of threat score, which can suppress the negative impact of sample imbalance on heavy precipitation forecasts by ignoring most of the events that are insensitive to heavy precipitation forecasts. Model correction experiments further indicate that the correction models trained with weighted cross entropy, differentiable threat score loss and Dice loss (WCE, DTS and Dice models) improve the precipitation forecasts of ECMWF (European Centre for Medium-Range Weather Forecasts) model. For the forecasts of all precipitation levels, Dice model effectively suppresses false alarms while improving hits, whereas WCE model greatly promotes false alarms while significantly improving hits. In contrast, DTS model significantly suppresses large precipitation false alarms while improving its hits, but causes substantial small precipitation false alarms while significantly promoting its hits. As a result, for large precipitation forecasts, both Dice and DTS models perform significantly better than WCE model, while for small precipitation forecasts, Dice model performs better than DTS and WCE models. Furthermore, the correction model combining ordinal classification and Dice loss exhibits the best forecasting skill. This model further suppresses the false alarms of large precipitation while improving its hits.

Multi-target segmentation of pancreas and pancreatic tumor based on fusion of attention mechanism

Existing neural network segmentation schemes perform well in the task of segmenting images of organs with large areas and clear morphology, such as the liver and lungs. However, it is difficult to segment organs with variable morphology and small target area, such as pancreas and tumors. In order to achieve accurate segmentation of pancreas and its cysts, MDAG-Net (Multi-dimensional Attention Gate Network) is proposed in this paper. Combining three attention mechanisms: spatial, channel and multi-dimensional feature map input, MDAG(Multi-dimensional Attention Gate) obtains the global distribution of semantic information in spatial and channel dimensions, filters redundant information in shallow feature maps, realizes feature response, and recalibrates convolution kernel parameters. In addition, the WML(Weighted cross entropy and MIoU loss function) loss can adaptively assign the weight of category loss and count the classification error of global pixels, which can increase the error attention of the target area and improving the segmentation accuracy of the network. The algorithm is experimented on the Task07_Pancreas dataset, compared with U-Net under the same conditions, the Dice coefficient, Precision, Recall rate and MIoU (Mean Intersection over Union) of MDAG-Net are improved by 5.3%, 1.5%, 12.7% and 7.6% respectively. The results show that MDAG-Net can accurately segment the region of pancreas and its cyst in CT(Computed Tomography) images, which proves that MDAG has better segmentation efficiency for such small target regions.

Comparison of Convolutional Neural Networks and Transformers for the Classification of Images of COVID-19, Pneumonia and Healthy Individuals as Observed with Computed Tomography.

In this work, the performance of five deep learning architectures in classifying COVID-19 in a multi-class set-up is evaluated. The classifiers were built on pretrained ResNet-50, ResNet-50r (with kernel size in the first convolutional layer), DenseNet-121, MobileNet-v3 and the state-of-the-art CaiT-24-XXS-224 (CaiT) transformer. The cross entropy and weighted cross entropy were minimised with Adam and AdamW. In total, 20 experiments were conducted with 10 repetitions and obtained the following metrics: accuracy (Acc), balanced accuracy (BA), F1 and F2 from the general F macro score, Matthew’s Correlation Coefficient (MCC), sensitivity (Sens) and specificity (Spec) followed by bootstrapping. The performance of the classifiers was compared by using the Friedman–Nemenyi test. The results show that less complex architectures such as ResNet-50, ResNet-50r and DenseNet-121 were able to achieve better generalization with rankings of 1.53, 1.71 and 3.05 for the Matthew Correlation Coefficient, respectively, while MobileNet-v3 and CaiT obtained rankings of 3.72 and 5.0, respectively.

Occupancy Grid Generation With Dynamic Obstacle Segmentation in Stereo Images

The detection of dynamic and static obstacles is a key task for the navigation of autonomous ground vehicles. The article presents a new algorithm for generating an occupancy map of the surrounding space from noisy point clouds obtained from one or several stereo cameras. The camera images are segmented by the proposed deep neural network FCN-ResNet-M-OC, which combines the speed of the FCN-ResNet method and improves the quality of the model using the concept of object context representation. The paper investigates supervised approaches to network training on unbalanced samples with road scenes such as the weighted cross entropy and the Focal Loss. The occupancy map is built from point clouds with semantic labels, in which static environment and potentially dynamic obstacles are highlighted. Our solution is operational in real time and applicable on platforms with limited computing resources. The approach was tested on autonomous vehicle datasets: Semantic KITTI, KITTI-360, Mapillary Vistas and custom OpenTaganrog. The usage of semantically labeled point clouds increased the precision of obstacle detection by an average of 17%. The performance of the entire approach on various computing platforms with Jetson Xavier, RTX3070, GPUs NVidia Tesla V100 is respectively from 10 to 15 FPS for input image resolution <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1920\times 1080$ </tex-math></inline-formula> pixels.

Cloud and snow detection of remote sensing images based on improved Unet3+

Cloud detection is an important step in remote sensing image processing and a prerequisite for subsequent analysis and interpretation of remote sensing images. Traditional cloud detection methods are difficult to accurately detect clouds and snow with very similar features such as color and texture. In this paper, the features of cloud and snow in remote sensing images are deeply extracted, and an accurate cloud and snow detection method is proposed based on the advantages of Unet3+ network in feature fusion. Firstly, color space conversion is performed on remote sensing images, RGB images and HIS images are used as input of Unet3+ network. Resnet 50 is used to replace the Unet3+ feature extraction network to extract remote sensing image features at a deeper level, and add the Convolutional Block Attention Module in Resnet50 to improve the network’s attention to cloud and snow. Finally, the weighted cross entropy loss is constructed to solve the problem of unbalanced sample number caused by high proportion of background area in the image. The results show that the proposed method has strong adaptability and moderate computation. The mPA value, mIoU value and mPrecision value can reach 92.76%, 81.74% and 86.49%, respectively. Compared with other algorithms, the proposed method can better eliminate all kinds of interference information in remote sensing images of different landforms and accurately detect cloud and snow in images.

Imbalanced Underwater Acoustic Target Recognition with Trigonometric Loss and Attention Mechanism Convolutional Network

A balanced dataset is generally beneficial to underwater acoustic target recognition. However, the imbalanced class distribution is always meted out in a real scene. To address this, a weighted cross entropy loss function based on trigonometric function is proposed. Then, the proposed loss function is applied in a multi-scale residual convolutional neural network (named MR-CNN-A network) embedded with an attention mechanism for the recognition task. Firstly, a multi-scale convolution kernel is used to obtain multi-scale features. Then, an attention mechanism is used to fuse these multi-scale feature maps. Furthermore, a cosx-function-weighted cross-entropy loss function is used to deal with the class imbalance in underwater acoustic data. This function adjusts the loss ratio of each sample by adjusting the loss interval of every mini-batch based on cosx term to achieve a balanced total loss for each class. Two imbalanced underwater acoustic data sets, ShipsEar and autonomous underwater vehicle (self-collected data) are used to evaluate the proposed network. The experimental results show that the proposed network outperforms the support vector machine and a simple convolutional neural network. Compared with the other three loss functions, the proposed loss function achieves better stability and adaptability. The results strongly demonstrate the validity of the proposed loss function and the network.

A multi-level feature-fusion-based approach to breast histopathological image classification

Previously, convolutional neural networks mostly used deep semantic feature information obtained from several convolutions for image classification. Such deep semantic features have a larger receptive field, and the features extracted are more effective as the number of convolutions increases, which helps in the classification of targets. However, this method tends to lose the shallow local features, such as the spatial connectivity and correlation of tumor region texture and edge contours in breast histopathology images, which leads to its recognition accuracy not being high enough. To address this problem, we propose a multi-level feature fusion method for breast histopathology image classification. First, we fuse shallow features and deep semantic features by attention mechanism and convolutions. Then, a new weighted cross entropy loss function is used to deal with the misjudgment of false negative and false positive. And finally, the correlation of spatial information is used to correct the misjudgment of some patches. We have conducted experiments on our own datasets and compared with the base network Inception-ResNet-v2, which has a high accuracy. The proposed method achieves an accuracy of 99.0% and an AUC of 99.9%.

Performance evaluation of deep transfer learning on multi-class identification of common weed species in cotton production systems

Precision weed management offers a promising solution for sustainable cropping systems through the use of chemical-reduced/non-chemical robotic weeding techniques, which apply suitable control tactics to individual weeds or small clusters. Therefore, accurate identification of weed species plays a crucial role in such systems to enable precise, individualized weed treatment. Despite recent progress, the development of a robust weed identification and localization system in the presence of unstructured field environments remains a serious challenge, requiring supervised modeling using large volumes of annotated data. This paper makes a first comprehensive evaluation of deep transfer learning (DTL) for identifying common weed species specific to cotton (Gossypium hirsutum L.) production systems in southern United States (U.S.). A new dataset for weed identification was created, consisting of 5187 color images of 15 weed classes collected under natural light conditions and at varied weed growth stages, in cotton fields (primarily in Mississippi and North Carolina) during the 2020 and 2021 growth seasons. We evaluated 35 state-of-the-art deep learning models through transfer learning with repeated holdout validations and established an extensive benchmark for the considered weed identification task. DTL achieved high classification accuracy of F1 scores exceeding 95%, requiring reasonably short training time (less than 2.5 h) across models. ResNeXt101 achieved the best overall F1-score of 98.93 ± 0.34%, whereas 10 out of the 35 models achieved F1 scores near or above 98.0%. However, the performance on minority weed classes with few training samples was less satisfactory for models trained with a conventional, unweighted cross entropy loss function. To address this issue, a weighted cross entropy loss function was adopted, which achieved substantially improved accuracies for minority weed classes (e.g., the F1-scores for Xception and MnasNet on the Spurred Anoda weed increased from 48% to 90% and 50% to 82%, respectively). Furthermore, a deep learning-based cosine similarity metric was employed to analyze the similarity among weed classes, assisting in the interpretation of classifications. Both the codes (https://github.com/Derekabc/CottonWeeds) for model benchmarking and the weed dataset (https://www.kaggle.com/yuzhenlu/cottonweedid15) of this study are made publicly available, which expect to be a valuable resource for future research on weed identification and beyond.

Time-Adaptive Transient Stability Assessment Based on the Gating Spatiotemporal Graph Neural Network and Gated Recurrent Unit

With the continuous expansion of the UHV AC/DC interconnection scale, online, high-precision, and fast transient stability assessment (TSA) is very important for the safe operation of power grids. In this study, a transient stability assessment method based on the gating spatiotemporal graph neural network (GSTGNN) is proposed. A time-adaptive method is used to improve the accuracy and speed of transient stability assessment. First, in order to reduce the impact of dynamic topology on TSA after fault removal, GSTGNN is used to extract and fuse the key features of topology and attribute information of adjacent nodes to learn the spatial data correlation and improve the evaluation accuracy. Then, the extracted features are input into the gated recurrent unit (GRU) to learn the correlation of data at each time. Fast and accurate evaluation results are output from the stability threshold. At the same time, in order to avoid the influence of the quality of training samples, an improved weighted cross entropy loss function with the K-nearest neighbor (KNN) idea is used to deal with the unbalanced training samples. Through the analysis of an example, it is proved from the data visualization that the TSA method can effectively improve the assessment accuracy and shorten the assessment time.

Sentinel-1-Based Water and Flood Mapping: Benchmarking Convolutional Neural Networks Against an Operational Rule-Based Processing Chain

In this study, the effectiveness of several convolutional neural network architectures (AlbuNet-34/FCN/DeepLabV3+/U-Net/U-Net++) for water and flood mapping using Sentinel-1 amplitude data is compared to an operational rule-based processor (S-1FS). This comparison is made using a globally distributed dataset of Sentinel-1 scenes and the corresponding ground truth water masks derived from Sentinel-2 data to evaluate the performance of the classifiers on a global scale in various environmental conditions. The impact of using single versus dual-polarized input data on the segmentation capabilities of AlbuNet-34 is evaluated. The weighted cross entropy loss is combined with the Lovász loss and various data augmentation methods are investigated. Furthermore, the concept of atrous spatial pyramid pooling used in DeepLabV3+ and the multiscale feature fusion inherent in U-Net++ are assessed. Finally, the generalization capacity of AlbuNet-34 is tested in a realistic flood mapping scenario by using additional data from two flood events and the Sen1Floods11 dataset. The model trained using dual polarized data outperforms the S-1FS significantly and increases the intersection over union (IoU) score by 5%. Using a weighted combination of the cross entropy and the Lovász loss increases the IoU score by another 2%. Geometric data augmentation degrades the performance while radiometric data augmentation leads to better testing results. FCN/DeepLabV3+/U-Net/U-Net++ perform not significantly different to AlbuNet-34. Models trained on data showing no distinct inundation perform very well in mapping the water extent during two flood events, reaching IoU scores of 0.96 and 0.94, respectively, and perform comparatively well on the Sen1Floods11 dataset.

Characteristics of Computed Tomography Images for Patients with Acute Liver Injury Caused by Sepsis under Deep Learning Algorithm

This study was aimed at exploring the application of image segmentation based on full convolutional neural network (FCN) in liver computed tomography (CT) image segmentation and analyzing the clinical features of acute liver injury caused by sepsis. The Sigmoid function, encoder-decoder, and weighted cross entropy loss function were introduced and optimized based on FCN. The Dice value, precision, recall rate, volume overlap error (VOE), relative volume difference (RVD), and root mean square error (RMSE) values of the optimized algorithms were compared and analyzed. 92 patients with sepsis were selected as the research objects, and they were divided into a nonacute liver injury group (50 cases) and acute liver injury group (42 cases) based on whether they had acute liver injury. The differences in the proportion of patients with different disease histories, the proportion of patients with different infection sites, the number of organ failure, and the time of admission to intensive care unit (ICU) were compared between the two groups. It was found that the optimized window CT image Dice value after preprocessing (0.704 ± 0.06) was significantly higher than the other two methods (P < 0.05). The Dice value, precision, and recall rate of the optimized-FCN algorithm were (0.826 ± 0.06), (0.91 ± 0.08), and (0.88 ± 0.09), respectively, which were significantly higher than other algorithms (P < 0.05). The VOE, RVD, and RMSE values were (21.19 ± 1.97), (10.45 ± 1.02), and (0.25 ± 0.02), respectively, which were significantly lower than other algorithms (P < 0.05). The proportion of patients with a history of drinking in the nonacute liver injury group was lower than that in the acute liver injury group (P < 0.05), and the proportion of patients with a history of hypotension was greatly higher than that in the nonacute liver injury group (P < 0.01). CT images of sepsis patients with acute liver injury showed that large areas of liver parenchyma mixed with high-density hematoma, the number of organ failures, and the length of stay in ICU were significantly higher than those in the nonacute liver injury group (P < 0.05). It showed that the optimization algorithm based on FCN greatly improved the performance of CT image segmentation. Long-term drinking, low blood pressure, number of organ failures, and length of stay in ICU were all related to sepsis and acute liver injury. Conclusion in this study could provide a reference basis for the diagnosis and prognosis of acute liver injury caused by sepsis.

CSF-Net: Color Spectrum Fusion Network for Semantic Labeling of Airborne Laser Scanning Point Cloud

Airborne laser scanning point cloud semantic labeling, which aims to identify the category of each point, plays a significant role in many applications, such as forest observing, powerline extraction, etc. Under the guidance of deep learning technology, the interpretation thought of point clouds has also greatly changed. However, owing to the irregular and unordered natures of point clouds, it is relatively difficult for classification model to distinguish some objects with similar geometry by single-modal data only. Fortunately, additional gain information, e.g., color spectrum which can be complementary to geometric information, is able to effectively promote the classification effect. Therefore, the design of fusion strategy is a critical part in model construction. In this article, aiming to capture more abstract semantic information for color spectrum data, we elaborate a color spectrum fusion (CSF) module. It can be flexibly integrated into a classification pipeline with just negligible parameters. Then, we expand data fusion thoughts for point clouds and color spectrum and investigate three possible fusion strategies. Accordingly, we develop three architectures to construct CSF-Nets. Ultimately, by taking a weighted cross entropy loss, we can train our CSF-Nets in an end-to-end manner. Experiments on two extensively used datasets: Vaihingen 3D and LASDU show that the presented three fusion approaches all can improve the performance, while the earlier fusion strategy performs the best. Besides, compared with other well-performed methods, CSF-Net is still able to achieve satisfactory performance on overall accuracy and m <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$F_{1}$</tex-math></inline-formula> -score indicator. This also validates the effectiveness of our multimodal fusion network.