High-resolution Network Research Articles

Mapping organizational complexity: a network-based approach to behavior through machine learning tools

The current uncertain, dynamic scenario calls for a systemic perspective when referring to organizational complexity and behavior. Our research contributes to the analysis of organizational complexity through multidimensional behavioral mapping. Our method uses machine learning tools to detect the interconnections between the different behaviors of a person in his/her operating context. First, the research project dealt with prototyping a model to read the organizational behavior, the related detection tool, and a data analysis methodology. It used machine learning tools and ended with a data visualization phase. We set our model to read the organizational behavior by comparing the literature benchmark theories with our field experience. The model was organized around 4 areas and 16 behaviors. These were the basis for singling out the indicators and the questionnaire items. The data analysis methodology aimed at detecting the interconnections between behaviors. We designed it by joining univariate analysis with a multivariate technique based on the application of machine learning tools. This led to a high-resolution network map through three specific steps: (a) creating a multidimensional topology based on a Kohonen Map (a type of unsupervised learning artificial neural network) to geometrically represent behavioral relationships; (b) implementing k-means clustering for identifying which areas of the map have behavior similarity or affinity factors; and (c) locating people and the various identified clusters within the map. The research highlighted the validity of machine learning tools to detect the multidimensionality of organizational behavior. Therefore, we could delineate the networking of the observed elements and visualize an otherwise unattainable complexity through multimedia and interactive reporting. Application in the field of research consisted of the design and development of a prototype integrated with our LMS platform via a plugin. Field experimentation confirmed the effectiveness of the method for creating professional growth and development paths. Furthermore, this experimentation allowed us to obtain significant data by applying our model to several sectors, namely pharmaceutical, TLC, banking, automotive, machinery, and services.

Automatic segmentation of the clinical target volume and organs at risk for rectal cancer radiotherapy using structure-contextual representations based on 3D high-resolution network

Radiotherapy is a cancer treatment that uses high doses of radiation to kill cancer cells. Segmentation of the clinical target volumes (CTVs) and organs at risk (OARs) is an essential step in rectal cancer radiotherapy treatment planning. However, the manual segmentation of CTVs and OARs is labor-intensive and prone to intra- and inter-rater variations. The challenge of this task lies in large shape variations and low-contrast structure boundary for some structures like intestine. The contextual information is very important to handle this challenge. Recently, deep learning has greatly improved the state-of-the-art results in automatic segmentation of anatomical structures including OARs and CTVs. However, existing approaches barely take advantages of relations among anatomical structures as contextual information, which limits the segmentation accuracy. In this study, we propose a novel structure-contextual representations approach based on 3D high-resolution network (3D HRNet-SCR) for segmentation of OARs and CTV of rectal cancer. First, we design a structure-contextual representation module (SCR) which could compute the representation for each structure of interest by aggregating the representations of voxels inside the structure and enhance the feature representation of each voxel with learned structure-contextual representations. Second, we extend the powerful 2D HRNet for semantic segmentation to 3D paradigm to better capture the contextual information across slices of volumetric data. Our propose approach integrates the proposed SCR module on top of the 3D HRNet to form a high-performance segmentation framework. Finally, we collect a large-scale rectal cancer dataset of 536 CT scans (total 64320 slices) for evaluation. Our proposed framework is extensively tested on this self-collected dataset, showing superiority compared with state-of-the-art OARs and CTV segmentation methods for rectal cancer treatment.

HR-MPF: high-resolution representation network with multi-scale progressive fusion for pulmonary nodule segmentation and classification

Accurate segmentation and classification of pulmonary nodules are of great significance to early detection and diagnosis of lung diseases, which can reduce the risk of developing lung cancer and improve patient survival rate. In this paper, we propose an effective network for pulmonary nodule segmentation and classification at one time based on adversarial training scheme. The segmentation network consists of a High-Resolution network with Multi-scale Progressive Fusion (HR-MPF) and a proposed Progressive Decoding Module (PDM) recovering final pixel-wise prediction results. Specifically, the proposed HR-MPF firstly incorporates boosted module to High-Resolution Network (HRNet) in a progressive feature fusion manner. In this case, feature communication is augmented among all levels in this high-resolution network. Then, downstream classification module would identify benign and malignant pulmonary nodules based on feature map from PDM. In the adversarial training scheme, a discriminator is set to optimize HR-MPF and PDM through back propagation. Meanwhile, a reasonably designed multi-task loss function optimizes performance of segmentation and classification overall. To improve the accuracy of boundary prediction crucial to nodule segmentation, a boundary consistency constraint is designed and incorporated in the segmentation loss function. Experiments on publicly available LUNA16 dataset show that the framework outperforms relevant advanced methods in quantitative evaluation and visual perception.

Wind Features Extracted from Weather Simulations for Wind-Wave Prediction Using High-Resolution Neural Networks

Nearshore wave forecasting is susceptible to changes in regional wind fields and environments. However, surface wind field changes are difficult to determine due to the lack of in situ observational data. Therefore, accurate wind and coastal wave forecasts during typhoon periods are necessary. The purpose of this study is to develop artificial intelligence (AI)-based techniques for forecasting wind–wave processes near coastal areas during typhoons. The proposed integrated models employ combined a numerical weather prediction (NWP) model and AI techniques, namely numerical (NUM)-AI-based wind–wave prediction models. This hybrid model comprising VGGNNet and High-Resolution Network (HRNet) was integrated with recurrent-based gated recurrent unit (GRU). Termed mVHR_GRU, this model was constructed using a convolutional layer for extracting features from spatial images with high-to-low resolution and a recurrent GRU model for time series prediction. To investigate the potential of mVHR_GRU for wind–wave prediction, VGGNet, HRNet, and Two-Step Wind-Wave Prediction (TSWP) were selected as benchmark models. The coastal waters in northeast Taiwan were the study area. The length of the forecast horizon was from 1 to 6 h. The mVHR_GRU model outperformed the HR_GRU, VGGNet, and TSWP models according to the error indicators. The coefficient of mVHR_GRU efficiency improved by 13% to 18% and by 13% to 15% at the Longdong and Guishandao buoys, respectively. In addition, in a comparison of the NUM–AI-based model and a numerical model simulating waves nearshore (SWAN), the SWAN model generated greater errors than the NUM–AI-based model. The results of the NUM–AI-based wind–wave prediction model were in favorable accordance with the observed results, indicating the feasibility of the established model in processing spatial data.

Semantic Segmentation Network for Surface Defect Detection of Automobile Wheel Hub Fusing High-Resolution Feature and Multi-Scale Feature

Surface defect detection of an automobile wheel hub is important to the automobile industry because these defects directly affect the safety and appearance of automobiles. At present, surface defect detection networks based on convolutional neural network use many pooling layers when extracting features, reducing the spatial resolution of features and preventing the accurate detection of the boundary of defects. On the basis of DeepLab v3+, we propose a semantic segmentation network for the surface defect detection of an automobile wheel hub. To solve the gridding effect of atrous convolution, the high-resolution network (HRNet) is used as the backbone network to extract high-resolution features, and the multi-scale features extracted by the Atrous Spatial Pyramid Pooling (ASPP) of DeepLab v3+ are superimposed. On the basis of the optical flow, we decouple the body and edge features of the defects to accurately detect the boundary of defects. Furthermore, in the upsampling process, a decoder can accurately obtain detection results by fusing the body, edge, and multi-scale features. We use supervised training to optimize these features. Experimental results on four defect datasets (i.e., wheels, magnetic tiles, fabrics, and welds) show that the proposed network has better F1 score, average precision, and intersection over union than SegNet, Unet, and DeepLab v3+, proving that the proposed network is effective for different defect detection scenarios.

High Frequential Resolution Networks: Considerations on a New Functional Brain Connectivity Framework.

Connectivity analyses are widely used to assess the interaction brain networks. This type of analyses is usually conducted considering the well-known classical frequency bands: delta, theta, alpha, beta, and gamma. However, this parcellation of the frequency content can bias the analyses, since it does not consider the between-subject variability or the particular idiosyncrasies of the connectivity patterns that occur within a band. In this study, we addressed these limitations by introducing the High Frequential Resolution Networks (HFRNs). HFRNs were constructed, using a narrow-bandwidth FIR bank filter of 1 Hz bandwidth, for two different connectivity metrics (Amplitude Envelope Correlation, AEC, and Phase Lag index, PLI) and for 3 different databases of MEG and EEG recordings. Results showed a noticeable similarity between the frequential evolution of PLI, AEC, and the Power Spectral Density (PSD) from MEG and EEG signals. Nonetheless, some technical remarks should be considered: (i) results at the gamma band should exclude the frequency range around 50 Hz due to abnormal connectivity patterns, consequence of the previously applied 50 Hz notch-filter; (ii) HFRNs patterns barely vary with the connection distance; and (iii) a low sampling frequency can exert a remarkable influence on HFRNs. To conclude, we proposed a new framework to perform connectivity analyses that allow to further analyze the frequency-based distribution of brain networks.

Whole brain segmentation with full volume neural network

Whole brain segmentation is an important neuroimaging task that segments the whole brain volume into anatomically labeled regions-of-interest. Convolutional neural networks have demonstrated good performance in this task. Existing solutions, usually segment the brain image by classifying the voxels, or labeling the slices or the sub-volumes separately. Their representation learning is based on parts of the whole volume whereas their labeling result is produced by aggregation of partial segmentation. Learning and inference with incomplete information could lead to sub-optimal final segmentation result. To address these issues, we propose to adopt a full volume framework, which feeds the full volume brain image into the segmentation network and directly outputs the segmentation result for the whole brain volume. The framework makes use of complete information in each volume and can be implemented easily. An effective instance in this framework is given subsequently. We adopt the 3D high-resolution network (HRNet) for learning spatially fine-grained representations and the mixed precision training scheme for memory-efficient training. Extensive experiment results on a publicly available 3D MRI brain dataset show that our proposed model advances the state-of-the-art methods in terms of segmentation performance.

Semantic Segmentation of Urban Buildings Using a High-Resolution Network (HRNet) with Channel and Spatial Attention Gates

In this study, building extraction in aerial images was performed using csAG-HRNet by applying HRNet-v2 in combination with channel and spatial attention gates. HRNet-v2 consists of transition and fusion processes based on subnetworks according to various resolutions. The channel and spatial attention gates were applied in the network to efficiently learn important features. A channel attention gate assigns weights in accordance with the importance of each channel, and a spatial attention gate assigns weights in accordance with the importance of each pixel position for the entire channel. In csAG-HRNet, csAG modules consisting of a channel attention gate and a spatial attention gate were applied to each subnetwork of stage and fusion modules in the HRNet-v2 network. In experiments using two datasets, it was confirmed that csAG-HRNet could minimize false detections based on the shapes of large buildings and small nonbuilding objects compared to existing deep learning models.

MommiNet-v2: Mammographic multi-view mass identification networks

Many existing approaches for mammogram analysis are based on single view. Some recent DNN-based multi-view approaches can perform either bilateral or ipsilateral analysis, while in practice, radiologists use both to achieve the best clinical outcome. MommiNet is the first DNN-based tri-view mass identification approach, which can simultaneously perform bilateral and ipsilateral analysis of mammographic images, and in turn, can fully emulate the radiologists’ reading practice. In this paper, we present MommiNet-v2, with improved network architecture and performance. Novel high-resolution network (HRNet)-based architectures are proposed to learn the symmetry and geometry constraints, to fully aggregate the information from all views for accurate mass detection. A multi-task learning scheme is adopted to incorporate both Breast Imaging-Reporting and Data System (BI-RADS) and biopsy information to train a mass malignancy classification network. Extensive experiments have been conducted on the public DDSM (Digital Database for Screening Mammography) dataset and our in-house dataset, and state-of-the-art results have been achieved in terms of mass detection accuracy. Satisfactory mass malignancy classification result has also been obtained on our in-house dataset.

Small-Target Complex-Scene Detection Method Based on Information Interworking High-Resolution Network.

The CNN (convolutional neural network)-based small target detection techniques for static complex scenes have been applied in many fields, but there are still certain technical challenges. This paper proposes a novel high-resolution small-target detection network named the IIHNet (information interworking high-resolution network) for complex scenes, which is based on information interworking processing technology. The proposed network not only can output a high-resolution presentation of a small target but can also keep the detection network simple and efficient. The key characteristic of the proposed network is that the target features are divided into three categories according to image resolution: high-resolution, medium-resolution, and low-resolution features. The basic features are extracted by convolution at the initial layer of the network. Then, convolution is carried out synchronously in the three resolution channels with information fusion in the horizontal and vertical directions of the network. At the same time, multiple utilizations and augmentations of feature information are carried out in the channel convolution direction. Experimental results show that the proposed network can achieve higher reasoning performance than the other compared networks without any compromise in terms of the detection effect.

Enhancing the estimation of fiber orientation distributions using convolutional neural networks

Local fiber orientation distributions (FODs) can be computed from diffusion magnetic resonance imaging (dMRI). The accuracy and ability of FODs to resolve complex fiber configurations benefits from acquisition protocols that sample a high number of gradient directions, a high maximum b-value, and multiple b-values. However, acquisition time and scanners that follow these standards are limited in clinical settings, often resulting in dMRI acquired at a single shell (single b-value). In this work, we learn improved FODs from clinically acquired dMRI. We evaluate patch-based 3D convolutional neural networks (CNNs) on their ability to regress multi-shell FODs from single-shell FODs, using constrained spherical deconvolution (CSD). We evaluate U-Net and High-Resolution Network (HighResNet) 3D CNN architectures on data from the Human Connectome Project and an in-house dataset. We evaluate how well each CNN can resolve FODs 1) when training and testing on datasets with the same dMRI acquisition protocol; 2) when testing on a dataset with a different dMRI acquisition protocol than used to train the CNN; and 3) when testing on a dataset with a fewer number of gradient directions than used to train the CNN. This work is a step towards more accurate FOD estimation in time- and resource-limited clinical environments.

Human pose estimation based on Improved High Resolution Network

Human pose estimation has become a hot problem in human-computer interaction and other intelligent application technologies, but the results obtained by previous deep learning networks are not very satisfactory when facing small-scale human instances. Therefore, in order to solve the problem of scale variation in human pose estimation, especially to pinpoint the keypoints of small-scale human instances, an improved High-Resolution Network (Improved HRNet) is proposed in this paper. The main improvement work in this paper is as follows: In this paper, a double attention mechanism is added in the forward transmission of the parallel sub-network, with the aim of assigning weights to the propagated information without changing the number of channels, assigning the information with high weights as useful information and reducing the interference caused by irrelevant information. In this paper, the network structure is validated using COCO dataset, and the average accuracy (Average Precision, AP) of Improved HRNet is 66.4, which is 2.3 higher than the average accuracy of High-Resolution Network (HRNet) with only 0.4% increase in parameters.

Classroom student posture recognition based on an improved high-resolution network

Due to the large number of students in a typical classroom and crowded seating, most features of student posture are often obscured, making it difficult to balance the accuracy in identifying student postures with computational efficiency. To solve this issue, a novel classroom student posture recognition method is proposed. First, to recognize the poses of multiple students in the classroom, we use the you-only-look-once (YOLOv3) algorithm for object detection and retrain it to detect human objects that are hunching on a table, creating the pose estimation network. Next, to improve the accuracy of the pose estimation network, we use the squeeze-and-excitation network structure that is embedded in the residual structure of high-resolution networks (HRNet). Finally, with the improved HRNet algorithm’s outputs of key human body points, we design a pose classification algorithm based on a support vector machine, to classify human poses in the classroom. Experiments show that the improved HRNet multi-person pose estimation algorithm yields the best mean average precision performance of 73.76% on the common objects in context (COCO) validation dataset. We further test the proposed algorithm on a customer dataset collected in a classroom and achieved a high recognition rate of 90.1% and good robustness.

Computerized assisted evaluation system for canine cardiomegaly via key points detection with deep learning

Cardiomegaly is the main imaging finding for canine heart diseases. There are many advances in the field of medical diagnosing based on imaging with deep learning for human being. However there are also increasing realization of the potential of using deep learning in veterinary medicine. We reported a clinically applicable assisted platform for diagnosing the canine cardiomegaly with deep learning. VHS (vertebral heart score) is a measuring method used for the heart size of a dog. The concrete value of VHS is calculated with the relative position of 16 key points detected by the system, and this result is then combined with VHS reference range of all dog breeds to assist in the evaluation of the canine cardiomegaly. We adopted HRNet (high resolution network) to detect 16 key points (12 and four key points located on vertebra and heart respectively) in 2274 lateral X-ray images (training and validation datasets) of dogs, the model was then used to detect the key points in external testing dataset (396 images), the AP (average performance) for key point detection reach 86.4 %. Then we applied an additional post processing procedure to correct the output of HRNets so that the AP reaches 90.9 %. This result signifies that this system can effectively assist the evaluation of canine cardiomegaly in a real clinical scenario.

Extraction of linear structures from digital terrain models using deep learning

Abstract. This paper explores the role deep convolutional neural networks play in automated extraction of linear structures using semantic segmentation techniques in Digital Terrain Models (DTMs). DTM is a regularly gridded raster created from laser scanning point clouds and represents elevations of the bare earth surface with respect to a reference. Recent advances in Deep Learning (DL) have made it possible to explore the use of semantic segmentation for detection of terrain structures in DTMs. This research examines two novel and practical deep convolutional neural network architectures i.e. an encoder-decoder network named as SegNet and the recent state-of-the-art high-resolution network (HRNet). This paper initially focuses on the pixel-wise binary classification in order to validate the applicability of the proposed approaches. The networks are trained to distinguish between points belonging to linear structures and those belonging to background. In the second step, multi-class segmentation is carried out on the same DTM dataset. The model is trained to not only detect a linear feature, but also to categorize it as one of the classes: hollow ways, roads, forest paths, historical paths, and streams. Results of the experiment in addition to the quantitative and qualitative analysis show the applicability of deep neural networks for detection of terrain structures in DTMs. From the deep learning models utilized, HRNet gives better results.

Automated COVID-19 Detection from Chest X-Ray Images: A High-Resolution Network (HRNet) Approach.

The pandemic, originated by novel coronavirus 2019 (COVID-19), continuing its devastating effect on the health, well-being, and economy of the global population. A critical step to restrain this pandemic is the early detection of COVID-19 in the human body to constraint the exposure and control the spread of the virus. Chest X-Rays are one of the non-invasive tools to detect this disease as the manual PCR diagnosis process is quite tedious and time-consuming. Our intensive background studies show that, the works till now are not efficient to produce an unbiased detection result. In this work, we proposed an automated COVID-19 classification method, utilizing available COVID and non-COVID X-Ray datasets, along with High-Resolution Network (HRNet) for feature extraction embedding with the UNet for segmentation purposes. To evaluate the proposed method, several baseline experiments have been performed employing numerous deep learning architectures. With extensive experiment, we got a significant result of 99.26% accuracy, 98.53% sensitivity, and 98.82% specificity with HRNet which surpasses the performances of the existing models. Finally, we conclude that our proposed methodology ensures unbiased high accuracy, which increases the probability of incorporating X-Ray images into the diagnosis of the disease.

Connectivity within regions characterizes epilepsy duration and treatment outcome.

Finding clear connectome biomarkers for temporal lobe epilepsy (TLE) patients, in particular at early disease stages, remains a challenge. Currently, the whole‐brain structural connectomes are analyzed based on coarse parcellations (up to 1,000 nodes). However, such global parcellation‐based connectomes may be unsuitable for detecting more localized changes in patients. Here, we use a high‐resolution network (~50,000‐nodes overall) to identify changes at the local level (within brain regions) and test its relation with duration and surgical outcome. Patients with TLE (n = 33) and age‐, sex‐matched healthy subjects (n = 36) underwent high‐resolution (~50,000 nodes) structural network construction based on deterministic tracking of diffusion tensor imaging. Nodes were allocated to 68 cortical regions according to the Desikan–Killany atlas. The connectivity within regions was then used to predict surgical outcome. MRI processing, network reconstruction, and visualization of network changes were integrated into the NICARA (https://nicara.eu). Lower clustering coefficient and higher edge density were found for local connectivity within regions in patients, but were absent for the global network between regions (68 cortical regions). Local connectivity changes, in terms of the number of changed regions and the magnitude of changes, increased with disease duration. Local connectivity yielded a better surgical outcome prediction (Mean value: 95.39% accuracy, 92.76% sensitivity, and 100% specificity) than global connectivity. Connectivity within regions, compared to structural connectivity between brain regions, can be a more efficient biomarker for epilepsy assessment and surgery outcome prediction of medically intractable TLE.

Deep High-Resolution Network for Low-Dose X-Ray CT Denoising

Low-dose computed tomography (LDCT) is clinically desirable because it reduces the radiation dose to patients. However, the quality of LDCT images is often suboptimal because of the inevitable strong quantum noise. Because of their unprecedented success in computer vision, deep learning (DL)-based techniques have been used for LDCT denoising. Despite DL models’ promising ability to remove noise, researchers have observed that the resolution of DL-denoised images is compromised, which decreases their clinical value. To mitigate this problem, in this work, we developed a more effective denoiser by introducing a high-resolution network (HRNet). HRNet consists of multiple branches of subnetworks that extract multiscale features that are fused together later, which substantially enhances the quality of the generated features and improves denoising performance. Experimental results demonstrated that the introduced HRNet-based denoiser outperformed the benchmarked U-Net–based denoiser, as it provided superior image resolution preservation and comparable, if not better, noise suppression. Quantitative evaluation in terms of root-mean-squared errors (RMSEs)/structure similarity index (SSIM) showed that the HRNet-based denoiser improve these values from 113.80/0.550 (LDCT) to 55.24/0.745 (HRNet), which outperformed the 59.87/0.712 for the U-Net–based denoiser.

Scale-aware attention-based multi-resolution representation for multi-person pose estimation

The performance of multi-person pose estimation has significantly improved with the development of deep convolutional neural networks. However, two challenging issues are still ignored but are key factors causing deterioration in the keypoint localization. These two issues are scale variation of human body parts and huge information loss caused by consecutive striding in multiple upsampling. In this paper, we present a novel network named ‘Scale-aware attention-based multi-resolution representation network’ (SaMr-Net) which targets to make the proposed method against scale variation and prevent the detail information loss in upsampling, leading more precisely keypoint estimation. The proposed architecture adopts the high-resolution network (HRNet) as the backbone, we first introduce dilated convolution into the backbone to expand the receptive field. Then, attention-based multi-scale feature fusion module is devised to modify the exchange units in the HRNet, allowing the network to learn the weights of each fusion component. Finally, we design a scale-aware keypoint regressor model that gradually integrates features from low to high resolution, enhancing the invariance in different scales of pose parts keypoint estimation. We demonstrate the superiority of the proposed algorithm over two benchmark datasets: (1) the MS COCO keypoint benchmark, and (2) the MPII human pose dataset. The comparison shows that our approach achieves superior results.

ICaps-Dfake: An Integrated Capsule-Based Model for Deepfake Image and Video Detection

Fake media is spreading like wildfire all over the internet as a result of the great advancement in deepfake creation tools and the huge interest researchers and corporations are showing to explore its limits. Now anyone can create manipulated unethical media forensics, defame, humiliate others or even scam them out of their money with a click of a button. In this research a new deepfake detection approach, iCaps-Dfake, is proposed that competes with state-of-the-art techniques of deepfake video detection and addresses their low generalization problem. Two feature extraction methods are combined, texture-based Local Binary Patterns (LBP) and Convolutional Neural Networks (CNN) based modified High-Resolution Network (HRNet), along with an application of capsule neural networks (CapsNets) implementing a concurrent routing technique. Experiments have been conducted on large benchmark datasets to evaluate the performance of the proposed model. Several performance metrics are applied and experimental results are analyzed. The proposed model was primarily trained and tested on the DeepFakeDetectionChallenge-Preview (DFDC-P) dataset then tested on Celeb-DF to examine its generalization capability. Experiments achieved an Area-Under Curve (AUC) score improvement of 20.25% over state-of-the-art models.