Candidate For Detection Research Articles

Evaluation of an automated intracranial aneurysm detection and rupture analysis approach using cascade detection and classification networks.

Intracranial aneurysm is commonly found in human brains especially for the elderly, and its rupture accounts for a high rate of subarachnoid hemorrhages. However, it is time-consuming and requires special expertise to pinpoint small aneurysms from computed tomography angiography (CTA) images. Deep learning-based detection has helped improve much efficiency but false-positives still render difficulty to be ruled out. To study the feasibility of deep learning algorithms for aneurysm analysis in clinical applications, this paper proposes a pipeline for aneurysm detection, segmentation, and rupture classification and validates its performance using CTA images of 1508 subjects. A cascade aneurysm detection model is employed by first using a fine-tuned feature pyramid network (FPN) for candidate detection and then applying a dual-channel ResNet aneurysm classifier to further reduce false positives. Detected aneurysms are then segmented by applying a traditional 3D V-Net to their image patches. Radiomics features of aneurysms are extracted after detection and segmentation. The machine-learning-based and deep learning-based rupture classification can be used to distinguish ruptured and un-ruptured ones. Experimental results show that the dual-channel ResNet aneurysm classifier utilizing image and vesselness information helps boost sensitivity of detection compared to single image channel input. Overall, the proposed pipeline can achieve a sensitivity of 90 % for 1 false positive per image, and 95 % for 2 false positives per image. For rupture classification the area under curve (AUC) of 0.906 can be achieved for the testing dataset. The results suggest feasibility of the pipeline for potential clinical use to assist radiologists in aneurysm detection and classification of ruptured and un-ruptured aneurysms.

A data-driven approach for ship-bridge collision candidate detection in bridge waterway

The consequences caused by bridge failures owing to the ship-bridge collision are always severe in terms of loss of life, economy, and environmental consequences to individuals and societies. The previous studies focused on the ship-bridge collision mainly concentrated on passive anti-collision, such as strengthening the bridge structure or setting anti-collision facilities. Compared with the previous research, the contribution of this work is to facilitate the reduction of collision risk of ship-bridge collision from the perspective of active anti-collision. A data-driven approach for ship-bridge collision candidate detection method in inland bridge waterways is proposed in this research. The approach is mainly divided into two steps: 1) The features (channel boundary, pier domain, and ship domain) of bridge waterways are identified using Kernel Density Estimation (KDE) method based on the historical AIS data; 2) Collision candidate detection with Velocity Obstacle (VO) method considering the identified features. This work can provide beneficial support for the ship-bridge active collision avoidance system.

Detection of astrophysical tau neutrino candidates in IceCube

High-energy tau neutrinos are rarely produced in atmospheric cosmic-ray showers or at cosmic particle accelerators, but are expected to emerge during neutrino propagation over cosmic distances due to flavor mixing. When high energy tau neutrinos interact inside the IceCube detector, two spatially separated energy depositions may be resolved, the first from the charged current interaction and the second from the tau lepton decay. We report a novel analysis of 7.5 years of IceCube data that identifies two candidate tau neutrinos among the 60 “High-Energy Starting Events” (HESE) collected during that period. The HESE sample offers high purity, all-sky sensitivity, and distinct observational signatures for each neutrino flavor, enabling a new measurement of the flavor composition. The measured astrophysical neutrino flavor composition is consistent with expectations, and an astrophysical tau neutrino flux is indicated at 2.8sigma significance.

Early detection of pine wilt disease tree candidates using time-series of spectral signatures.

Pine wilt disease (PWD), caused by pine wood nematode (PWN), poses a tremendous threat to global pine forests because it can result in rapid and widespread infestations within months, leading to large-scale tree mortality. Therefore, the implementation of preventive measures relies on early detection of PWD. Unmanned aerial vehicle (UAV)-based hyperspectral images (HSI) can detect tree-level changes and are thus an effective tool for forest change detection. However, previous studies mainly used single-date UAV-based HSI data, which could not monitor the temporal changes of disease distribution and determine the optimal detection period. To achieve these purposes, multi-temporal data is required. In this study, Pinus koraiensis stands were surveyed in the field from May to October during an outbreak of PWD. Concurrently, multi-temporal UAV-based red, green, and blue bands (RGB) and HSI data were also obtained. During the survey, 59 trees were confirmed to be infested with PWD, and 59 non-infested trees were used as control. Spectral features of each tree crown, such as spectral reflectance, first and second-order spectral derivatives, and vegetation indices (VIs), were analyzed to identify those useful for early monitoring of PWD. The Random Forest (RF) classification algorithm was used to examine the separability between the two groups of trees (control and infested trees). The results showed that: (1) the responses of the tree crown spectral features to PWD infestation could be detected before symptoms were noticeable in RGB data and field surveys; (2) the spectral derivatives were the most discriminable variables, followed by spectral reflectance and VIs; (3) based on the HSI data from July to October, the two groups of trees were successfully separated using the RF classifier, with an overall classification accuracy of 0.75–0.95. Our results illustrate the potential of UAV-based HSI for PWD early monitoring.

Automatic detection of primary and metastatic lesions on cervicothoracic region and whole-body bone using a uniform machine-learnable approach for [18F]-FDG-PET/CT image analysis

We propose a method to detect primary and metastatic lesions with Fluorine−18 fluorodeoxyglucose (FDG) accumulation in the lung field, neck, mediastinum, and bony regions on the FDG-PET/CT images. To search for systemic lesions, various anatomical structures must be considered. The proposed method is addressed by using an extraction process for anatomical regions and a uniform lesion detection approach. The uniform approach does not utilize processes that reflect any region-specific anatomical aspects but has a machine-learnable framework. Therefore, it can work as a lesion detection process for a specific anatomical region if it machine-learns the specific region data. In this study, three lesion detection processes for the whole-body bone region, lung field, or neck-mediastinum region are obtained. These detection processes include lesion candidate detection and false positive (FP) candidate elimination. The lesion candidate detection is based on a voxel anomaly detection with a one-class support vector machine. The FP candidate elimination is performed using an AdaBoost classifier ensemble. The image features used by the ensemble are selected sequentially during training and are optimal for candidate classification. Three-fold cross-validation was used to detect performance with the 54 diseased FDG-PET/CT images. The mean sensitivity for detecting primary and metastatic lesions at 3 FPs per case was 0.89 with a 0.10 standard deviation (SD) in the bone region, 0.80 with a 0.10 SD in the lung field, and 0.87 with a 0.10 SD in the neck region. The average areas under the ROC curve were 0.887 with a 0.125 SD for detecting bone metastases, 0.900 with a 0.063 SD for detecting pulmonary lesions, and 0.927 with a 0.035 SD for detecting the neck-mediastinum lesions. These detection performances indicate that the proposed method could be applied clinically. These results also show that the uniform approach has high versatility for providing various lesion detection processes.

Microaneurysms detection in retinal images using a multi-scale approach

One of the earliest signs of diabetic retinopathy is the appearance of small and round red dots in retinal fundus images, known as microaneurysms. Microaneurysms automatic detection is a challenging task due to their small and subtle nature. In this paper, a new multi-scale algorithm developed for the automatic detection of microaneurysms in retinal fundus images is proposed. The proposed method is divided into three stages, namely, image pre-processing, microaneurysm candidates detection and retinal vessels detection, and labelling of the microaneurysm candidates. Apart the initial image pre-processing, all stages of the algorithm are designed using a multi-scale framework, providing the mechanism for the effective microaneurysms detection. Four publicly available databases are used for performance evaluation. The proposed method proves to be competitive against other state-of-the-art approaches in microaneurysms detection and diabetic retinopathy detection.

Exploratory analysis of text duplication in peer-review reveals peer-review fraud and paper mills

Comments received from referees during peer-review were analysed to determine the rates of duplication and partial duplication. It is very unusual for 2 different referees to submit identical comments, so the rare cases where this happens are of interest. In some cases, it appears that paper-mills create fake referee accounts and use them to submit fake peer-review reports. These include comments that are copied and pasted across multiple reviews. Searching for duplication in referee comments is therefore an effective method to search for misconduct generally, since the forms of misconduct committed by paper-mills go beyond peer-review fraud. These search methods allow the automatic detection of misconduct candidates which may then be investigated carefully to confirm if misconduct has indeed taken place. There are innocent reasons why referees might share template reports, so these methods are not intended to automatically diagnose misconduct.

A novel lung nodule accurate detection of computerized tomography images based on convolutional neural network and probability graph model

AbstractPrecisely detecting lung nodules from original computerized tomography (CT) images is a critical technology in the earlier screening of lung cancer. Therefore, the domain of accurate detection has gradually attracted the attention of researchers. However, due to the complex characteristics of pulmonary nodules and the limitations of CT imaging property, detecting nodules with high accuracy from lung CT images is a challenging task. This article proposes an effective and robust detection network to accurately detect lung nodules by innovatively implementing a probability graph model in the candidate detection and false‐positive reduction phase. Different from previous works which use complex 3‐dimensional image information to reduce false positives, we propose two effective probability graph mechanisms, which analyze multiscale information and continuous slices (interslice changes) motion information to improve performance. We evaluated our method on an open‐source LIDC‐IDRI dataset which contains a total of 243,958 CT images and achieved high‐precision lung nodule detection results (sensitivity score of 0.945). Via introducing multiscale information and the dynamic information of the interslice, the task of lung nodule detection obtains higher precision detection results than other similar methods.

Современные подходы к выявлению скрываемой информации

the article is devoted to an urgent problem - the identification of hidden information. Many scientific studies of domestic and foreign authors reveal this topic in order to provide practical assistance in the framework of crime detection and personnel selection of candidates for service. However, research in this area is still ongoing, since a universal method has not yet been found that takes into account all kinds of informative sources of information received. The topic of detecting lies in statements during investigative activities, personnel selection of candidates for service remains not fully disclosed and requires further research. The article analyzes the existing modern methods of lie recognition, both instrumental and without the use of special equipment, and compares the reliability of the information received. Based on the study and generalization of existing modern methods, the author comes to the conclusion that combined methods of diagnosing false information are most effective when the equipment is auxiliary, complementing the main method and serves to fix the processes taking place

Size-adaptive mediastinal multilesion detection in chest CT images via deep learning and a benchmark dataset.

Many deep learning methods have been developed for pulmonary lesion detection in chest computed tomography (CT) images. However, these methods generally target one particular lesion type, that is, pulmonary nodules. In this work, we intend to develop and evaluate a novel deep learning method for a more challenging task, detecting various benign and malignant mediastinal lesions with wide variations in sizes, shapes, intensities, and locations in chest CT images. Our method for mediastinal lesion detection contains two main stages: (a) size-adaptive lesion candidate detection followed by (b) false-positive (FP) reduction and benign-malignant classification. For candidate detection, an anchor-free and one-stage detector, namely 3D-CenterNet is designed to locate suspicious regions (i.e., candidates with various sizes) within the mediastinum. Then, a 3D-SEResNet-based classifier is used to differentiate FPs, benign lesions, and malignant lesions from the candidates. We evaluate the proposed method by conducting five-fold cross-validation on a relatively large-scale dataset, which consists of data collected on 1136 patients from a grade A tertiary hospital. The method can achieve sensitivity scores of 84.3% ± 1.9%, 90.2% ± 1.4%, 93.2% ± 0.8%, and 93.9% ± 1.1%, respectively, in finding all benign and malignant lesions at 1/8, 1/4, ½, and 1 FPs per scan, and the accuracy of benign-malignant classification can reach up to 78.7% ± 2.5%. The proposed method can effectively detect mediastinal lesions with various sizes, shapes, and locations in chest CT images. It can be integrated into most existing pulmonary lesion detection systems to promote their clinical applications. The method can also be readily extended to other similar 3D lesion detection tasks.

Pre-Clinical In-Vitro Studies on Parameters Governing Immune Complex Formation.

The success of biotherapeutics is often challenged by the undesirable events of immunogenicity in patients, characterized by the formation of anti-drug antibodies (ADA). Under specific conditions, the ADAs recognizing the biotherapeutic can trigger the formation of immune complexes (ICs), followed by cascades of subsequent effects on various cell types. Hereby, the connection between the characteristics of ICs and their downstream impact is still not well understood. Factors governing the formation of ICs and the characteristics of these IC species were assessed systematically in vitro. Classic analytical methodologies such as SEC-MALS and SV-AUC, and the state-of-the-art technology mass photometry were applied for the characterization. The study demonstrates a clear interplay between (1) the absolute concentration of the involved components, (2) their molar ratios, (3) structural features of the biologic, (4) and of its endogenous target. This surrogate study design and the associated analytical tool-box is readily applicable to most biotherapeutics and provides valuable insights into mechanisms of IC formation prior to FIH studies. The applicability is versatile—from the detection of candidates with immunogenicity risks during developability assessment to evaluation of the impact of degraded or post-translationally modified biotherapeutics on the formation of ICs.

Augmented networks for faster brain metastases detection in T1-weighted contrast-enhanced 3D MRI.

Early detection of brain metastases (BM) is one of the determining factors for the successful treatment of patients with cancer; however, the accurate detection of small BM lesions (< 15mm) remains a challenging task. We previously described a framework for the detection of small BM in single-sequence gadolinium-enhanced T1-weighted 3D MRI datasets. It combined classical image processing (IP) with a dedicated convolutional neural network, taking approximately 30s to process each dataset due to computation-intensive IP stages. To overcome the speed limitation, this study aims to reformulate the framework via an augmented pair of CNNs (eliminating the IP) to reduce the processing times while preserving the BM detection performance. Our previous implementation of the BM detection algorithm utilized Laplacian of Gaussians (LoG) for the candidate selection portion of the solution. In this study, we introduce a novel BM candidate detection CNN (cdCNN) to replace this classical IP stage. The network is formulated to have (1) a similar receptive field as the LoG method, and (2) a bias for the detection of BM lesion loci. The proposed CNN is later augmented with a classification CNN to perform the BM detection task. The cdCNN achieved 97.4% BM detection sensitivity when producing 60K candidates per 3D MRI dataset, while the LoG achieved 96.5% detection sensitivity with 73K candidates. The augmented BM detection framework generated on average 9.20 false-positive BM detections per patient for 90% sensitivity, which is comparable with our previous results. However, it processes each 3D data in 1.9s, presenting a 93.5% reduction in the computation time.

TnFLXopen: Markerless Transposons for Functional Fluorescent Fusion Proteins and Protein Interaction Prediction.

ABSTRACTFluorescence microscopy of cells expressing proteins translationally linked to a fluorophore can be a powerful tool to investigate protein localization dynamics in vivo. One major obstacle to reliably analyze biologically relevant localization is the construction of a fusion protein that is both fluorescent and functional. Here, we develop a strategy to construct fluorescent fusions at theoretically any location in the protein by using TnFLXopen random transposon mutagenesis to randomly insert a gene encoding a fluorescent protein. Moreover, insertions within a target gene are enriched by an inducible gene-trap strategy and selection by fluorescence activated cell sorting. Using this approach, we isolate a variety of fluorescent fusions to FtsZ that exhibit ring-like localization and a fusion to the flagellar stator protein that both is functional for supporting motility and localizes as fluorescent puncta. Finally, we further modify TnFLXopen to insert the coding sequence for the C-terminal half of mVenus for use in bimolecular fluorescence complementation (BiFC) and the in vivo detection of protein-protein interaction candidates. As proof-of-concept, the DivIVA polar scaffolding protein was fused to the N terminus of mVenus, the C terminus of mVenus was delivered by transposition, and a combination of fluorescence activated cell sorter (FACS) sorting and whole-genome sequencing identified the known self-interaction of DivIVA as well as other possible candidate interactors. We suggest that the FACS selection is a viable alternative to antibiotic selection in transposon mutagenesis that can generate new fluorescent tools for in vivo protein characterization.IMPORTANCE Transposon mutagenesis is a powerful tool for random mutagenesis, as insertion of a transposon and accompanying antibiotic resistance cassette often disrupt gene function. Here, we present a series of transposons with fluorescent protein genes which, when integrated in frame, may be selected with a fluorescence activated cell sorter (FACS). An open reading frame runs continuously through the transposon such that fluorescent protein fusions may be inserted theoretically anywhere in the primary sequence and potentially preserve function of the target protein. Finally, the transposons were further modified to randomly insert a partial fluorescent protein compatible with bimolecular fluorescence complementation (BiFC) to identify protein interaction candidates.

STURE: Spatial–Temporal Mutual Representation Learning for robust data association in online multi-object tracking

Online multi-object tracking (MOT) is a longstanding task for computer vision and intelligent vehicle platform. At present, the main paradigm is tracking-by-detection, and the main difficulty of this paradigm is how to associate current candidate detections with historical tracklets. However, in the MOT scenarios, each historical tracklet is composed of an object sequence, while each candidate detection is just a flat image, which lacks temporal features of the object sequence. The feature difference between current candidate detections and historical tracklets makes the object association much harder. Therefore, we propose a Spatial–Temporal Mutual Representation Learning (STURE) approach which learns spatial–temporal representations between current candidate detections and historical sequences in a mutual representation space. For historical tracklets, the detection learning network is forced to match the representations of sequence learning network in a mutual representation space. The proposed approach is capable of extracting more distinguishing detection and sequence representations by using various designed losses in object association. As a result, spatial–temporal feature is learned mutually to reinforce the current detection features, and the feature difference can be relieved. To prove the robustness of the STURE, it is applied to the public MOT challenge benchmarks and performs well compared with various state-of-the-art online MOT trackers based on identity-preserving metrics.

Oil Spill Candidate Detection Using a Conditional Random Field Model on Simulated Compact Polarimetric Imagery

Although the compact polarimetric (CP) synthetic aperture radar (SAR) mode of the RADARSAT Constellation Mission (RCM) offers new opportunities for oil spill candidate detection, there has not been an efficient machine learning model explicitly designed to utilize this new CP SAR data for improved detection. This paper presents a conditional random field model based on the Wishart mixture model (CRF-WMM) to detect oil spill candidates in CP SAR imagery. First, a “Wishart mixture model” (WMM) is designed as the unary potential in the CRF-WMM to address the class-dependent information of oil spill candidates and oil free water. Second, we introduce a new similarity measure based on CP statistics designed as a pairwise potential in the CRF-WMM model so that pixels with strong spatial connections have the same class label. Finally, we investigate three different optimization approaches to solve the resulting maximum a posterior (MAP) problem, namely iterated conditional modes (ICM), simulated annealing (SA), and graph cuts (GC). The results show that our proposed CRF-WMM model can delineate oil spill candidates better than the traditional CRF approaches, and that the GC algorithm provides the best optimization.

Automatic microchannel detection using deep learning in intravascular optical coherence tomography images.

Microchannel formation is known to be a significant marker of plaque vulnerability, plaque rupture, and intraplaque hemorrhage, which are responsible for plaque progression. We developed a fully-automated method for detecting microchannels in intravascular optical coherence tomography (IVOCT) images using deep learning. A total of 3,075 IVOCT image frames across 41 patients having 62 microchannel segments were analyzed. Microchannel was manually annotated by expert cardiologists, according to previously established criteria. In order to improve segmentation performance, pre-processing including guidewire detection/removal, lumen segmentation, pixel-shifting, and noise filtering was applied to the raw (r,θ) IVOCT image. We used the DeepLab-v3 plus deep learning model with the Xception backbone network for identifying microchannel candidates. After microchannel candidate detection, each candidate was classified as either microchannel or no-microchannel using a convolutional neural network (CNN) classification model. Our method provided excellent segmentation of microchannel with a Dice coefficient of 0.811, sensitivity of 92.4%, and specificity of 99.9%. We found that pre-processing and data augmentation were very important to improve results. In addition, a CNN classification step was also helpful to rule out false positives. Furthermore, automated analysis missed only 3% of frames having microchannels and showed no false positives. Our method has great potential to enable highly automated, objective, repeatable, and comprehensive evaluations of vulnerable plaques and treatments. We believe that this method is promising for both research and clinical applications.

A Channel to Form Fast-spinning Black Hole–Neutron Star Binary Mergers as Multimessenger Sources

After the successful detection of a gravitational-wave (GW) signal and its associated electromagnetic (EM) counterparts from GW170817, neutron star–black hole (NSBH) mergers have been highly expected to be the next type of multimessenger source. However, despite the detection of several NSBH merger candidates during the GW third observation run, no confirmed EM counterparts from these sources have been identified. The most plausible explanation is that these NSBH merger candidates were plunging events mainly because the primary black holes (BHs) had near-zero projected aligned spins based on GW observations. In view of the fact that neutron stars (NSs) can be easily tidally disrupted by BHs with high projected aligned spins, we study an evolution channel to form NSBH binaries with fast-spinning BHs, the properties of BH mass and spin, and their associated tidal disruption probability. We find that if the NSs are born first, the companion helium stars would be tidally spun up efficiently, and would thus finally form fast-spinning BHs. If BHs do not receive significant natal kicks at birth, these NSBH binaries that can merge within Hubble time would have BHs with projected aligned spins χ z ≳ 0.8 and, hence, can certainly allow tidal disruption to happen. Even if significant BH kicks are considered for a small fraction of NSBH binaries, the projected aligned spins of BHs are χ z ≳ 0.2. These systems can still be disrupted events unless the NSs are very massive. Thus, NS-first-born NSBH mergers would be promising multimessenger sources. We discuss various potential EM counterparts associated with these systems and their detectability in the upcoming fourth observation run.

Hepatitis B and C screening in hospitalized patients with SARS-CoV-2 infection

AimsTo evaluate the results of a hepatitis B and C screening program in hospitalized COVID-19 patients.MethodTransversal prospective study conducted in two Spanish hospitals. Patients admitted from March 1st to December 31st 2020 with a diagnosis of COVID-19 were tested for markers of hepatitis B (HBsAg, anti-HBc) and C (anti-HCV, HCV RNA) infection.ResultsIn this period, 4662 patients with COVID-19 were admitted to our centers: 56.3% were male, median age was 76 (0–104) years. Data regarding HBV infection was available in 2915 (62.5%) patients; 253 (8.75%) were anti-HBc + and 11 (0.38%) HBsAg+. From these, 4 patients did not have a previous diagnosis of hepatitis B, 7 received corticosteroids and one received prophylaxis. There was one HBV reactivation. Anti-HCV were available in 2895 (62%) patients; 24 (0.83%) were positive. From these, 13 patients had a previous hepatitis C diagnosis: 10 patients had been treated with SVR, one achieved spontaneous cure and 2 did not receive treatment. From the 11 previously unknown anti-VHC + patients, 10 had a negative HCV RNA. Overall, only 3 (0.10%) patients tested RNA HCV + . However, none received HCV treatment (2 older than 90 years with comorbidities, 1 died from COVID-19).ConclusionScreening of hepatitis C infection in hospitalized COVID-19 patients seems less useful than expected. The low prevalence of active infection after antiviral treatments and the high age of our population limit the detection of potential candidates for treatment. HBV screening should be aimed to prevent reactivation under immunosuppressive treatments.

Automatic coronary plaque detection, classification, and stenosis grading using deep learning and radiomics on computed tomography angiography images: a multi-center multi-vendor study.

An automatic system utilizing both the advantages of the neural network and the radiomics was proposed for coronary plaque detection, classification, and stenosis grading. This study retrospectively included 505 patients with 127,763 computed tomography angiography (CTA) images from 5 medical center. A convolutional neural network (CNN) model was used to segment the coronary artery, detect the plaque candidate, and extract the image patch with high computation efficiency. The manually designed radiomics feature extractor was utilized to collect plaque patterns, followed by the different classifiers to perform the plaque classification and stenosis grading. The CNN model achieved 100% of sensitivity and the highest positive predictive value (83.9%) than U-Net and baseline model in plaque candidate detection. Twenty-six representative radiomics features were selected to construct the classifiers. Among different models, the gradient-boosting decision tree (GBDT) achieved the best performance in plaque classification (accuracy: 87.0%, sensitivity: 83.2%, specificity: 91.4%) and stenosis grading (accuracy: 90.9%, sensitivity: 84.1%, specificity: 95.7%). GBDT also achieved the highest AUC of 0.873 in plaque classification and 0.910 in stenosis grading. The computation time of processing one patient was 56.2 ± 5.7 s which was significantly less than radiologist manual analysis (285.6 ± 134.5 s, p = 0.0001). In this study, an automatic workflow was proposed to detect and analyze coronary plaques with high accuracy and efficiency, showing the potential in clinical application. • The proposed automatic system integrated deep learning and radiomics to perform the coronary plaque analysis. • The proposed automatic system achieved high accuracy in both plaque classification and stenosis grading. • The proposed automatic system was five times more efficient than radiologist manual analysis.

GKNet: Grasp keypoint network for grasp candidates detection

Contemporary grasp detection approaches employ deep learning to achieve robustness to sensor and object model uncertainty. The two dominant approaches design either grasp-quality scoring or anchor-based grasp recognition networks. This paper presents a different approach to grasp detection by treating it as keypoint detection in image-space. The deep network detects each grasp candidate as a pair of keypoints, convertible to the grasp representation g = {x,y,w,θ} T , rather than a triplet or quartet of corner points. Decreasing the detection difficulty by grouping keypoints into pairs boosts performance. To promote capturing dependencies between keypoints, a non-local module is incorporated into the network design. A final filtering strategy based on discrete and continuous orientation prediction removes false correspondences and further improves grasp detection performance. GKNet, the approach presented here, achieves a good balance between accuracy and speed on the Cornell and the abridged Jacquard datasets (96.9% and 98.39% at 41.67 and 23.26 fps). Follow-up experiments on a manipulator evaluate GKNet using four types of grasping experiments reflecting different nuisance sources: static grasping, dynamic grasping, grasping at varied camera angles, and bin picking. GKNet outperforms reference baselines in static and dynamic grasping experiments while showing robustness to varied camera viewpoints and moderate clutter. The results confirm the hypothesis that grasp keypoints are an effective output representation for deep grasp networks that provide robustness to expected nuisance factors.