Search

A point cloud is a simple and concise 3D representation, but point cloud generation is a long-term challenging task in 3D vision. However, most existing methods only focus on their effectiveness of generation and auto-encoding separately. Furthermore, both generative adversarial networks (GANs) and auto-encoders (AEs) are the most popular generative models. But there is a lack of related research that investigates the implicit connections between them in the field of point cloud generation. Thus, we propose a new bidirectional network (BI-Net) trained with collaborative learning, introducing more priors through the alternate parameter optimizations of a GAN and AE combination, which is different from the way of combining them at the network structure and loss function level. Specifically, BI-Net acts as a GAN and AE in different data processing directions, where their network structures can be reused. If optimizing only the GAN without the AE, there is no direct constraint of ground truth on the generator's parameter optimization. This unique approach enables better network optimization and leads to superior generation results. Moreover, we propose a nearest neighbor mutual exclusion (NNME) loss to further homogenize the spatial distribution of generated points during the reverse direction. Extensive experiments were conducted, and the results show that the BI-Net produces competitive and high-quality results on reasonable structure and uniform distributions compared to existing state-of-the-art methods. We believe that our network structure (BI-Net) with collaborative learning could provide a new promising method for future point cloud generation tasks. [ABSTRACT FROM AUTHOR]

It is significant to identify the moldy status of stored maize by fungi infection in the early stage. Hyperspectral imaging (HSI) combined with the sparse auto-encoders (SAE) and convolutional neural network (CNN) algorithms was used to classify the moldy grades of maize kernels. The HSI data were obtained in the range of 400–1000 nm, and four grades from health to heavy mildew were distinguished using the measured fungal spores of maize. The depth spectral features were represented using SAE and the image features were extracted by CNN. K nearest neighbors, support vector machine (SVM), and partial least squares discriminant analysis classifiers were combined with the spectral and image features to establish classification models to identify the different moldy grades of maize kernels. The comparison results indicated that the fusion of SAE and CNN combined with the SVM classifier to construct the SAE-CNN-SVM model had the most satisfactory identification result with high correct recognition rates of 99.47% and 98.94% for the training and testing sets, respectively, and the values of sensitivity and specificity were 0.95–1. The moldy grades were presented intuitively on the maize image based on pixels or kernel-wise. Therefore, the HSI with the SAE-CNN-SVM model had good recognition ability for the early detection of moldy maize kernels, which could potentially provide technical support for the development of online detection of moldy maize kernels during storage. [ABSTRACT FROM AUTHOR]

Active learning is a unique abstraction of machine learning techniques where the model/algorithm could guide users for annotation of a set of data points that would be beneficial to the model, unlike passive machine learning. The primary advantage being that active learning frameworks select data points that can accelerate the learning process of a model and can reduce the amount of data needed to achieve full accuracy as compared to a model trained on a randomly acquired data set. Multiple frameworks for active learning combined with deep learning have been proposed, and the majority of them are dedicated to classification tasks. Herein, we explore active learning for the task of segmentation of medical imaging data sets. We investigate our proposed framework using two datasets: 1.) MRI scans of the hippocampus, 2.) CT scans of pancreas and tumors. This work presents a query-by-committee approach for active learning where a joint optimizer is used for the committee. At the same time, we propose three new strategies for active learning: 1.) increasing frequency of uncertain data to bias the training data set; 2.) Using mutual information among the input images as a regularizer for acquisition to ensure diversity in the training dataset; 3.) adaptation of Dice log-likelihood for Stein variational gradient descent (SVGD). The results indicate an improvement in terms of data reduction by achieving full accuracy while only using 22.69% and 48.85% of the available data for each dataset, respectively. [ABSTRACT FROM AUTHOR]

Performing deep neural network (DNN) inference in real time requires excessive network resources, which poses a big challenge to the resource-limited industrial Internet of things (IIoT) networks. To address the challenge, in this paper, we introduce an end-edge-cloud orchestration architecture, in which the inference task assignment and DNN model placement are flexibly coordinated. Specifically, the DNN models, trained and pre-stored in the cloud, are properly placed at the end and edge to perform DNN inference. To achieve efficient DNN inference, a multi-dimensional resource management problem is formulated to maximize the average inference accuracy while satisfying the strict delay requirements of inference tasks. Due to the mix-integer decision variables, it is difficult to solve the formulated problem directly. Thus, we transform the formulated problem into a Markov decision process which can be solved efficiently. Furthermore, a deep reinforcement learning based resource management scheme is proposed to make real-time optimal resource allocation decisions. Simulation results are provided to demonstrate that the proposed scheme can efficiently allocate the available spectrum, caching, and computing resources, and improve average inference accuracy by 31.4 % compared with the deep deterministic policy gradient benchmark. [ABSTRACT FROM AUTHOR]

Background: Pneumonia is the most frequently encountered postoperative pulmonary complications (PPC) after orthotopic liver transplantation (OLT), which cause high morbidity and mortality rates. We aimed to develop a model to predict postoperative pneumonia in OLT patients using machine learning (ML) methods.Methods: Data of 786 adult patients underwent OLT at the Third Affiliated Hospital of Sun Yat-sen University from January 2015 to September 2019 was retrospectively extracted from electronic medical records and randomly subdivided into a training set and a testing set. With the training set, six ML models including logistic regression (LR), support vector machine (SVM), random forest (RF), adaptive boosting (AdaBoost), extreme gradient boosting (XGBoost) and gradient boosting machine (GBM) were developed. These models were assessed by the area under curve (AUC) of receiver operating characteristic on the testing set. The related risk factors and outcomes of pneumonia were also probed based on the chosen model.Results: 591 OLT patients were eventually included and 253 (42.81%) were diagnosed with postoperative pneumonia, which was associated with increased postoperative hospitalization and mortality (P < 0.05). Among the six ML models, XGBoost model performed best. The AUC of XGBoost model on the testing set was 0.734 (sensitivity: 52.6%; specificity: 77.5%). Pneumonia was notably associated with 14 items features: INR, HCT, PLT, ALB, ALT, FIB, WBC, PT, serum Na+, TBIL, anesthesia time, preoperative length of stay, total fluid transfusion and operation time.Conclusion: Our study firstly demonstrated that the XGBoost model with 14 common variables might predict postoperative pneumonia in OLT patients. [ABSTRACT FROM AUTHOR]

Echocardiography is the first-line diagnostic technique for heart diseases. Although artificial intelligence techniques have made great improvements in the analysis of echocardiography, the major limitations remain to be the built neural networks are normally adapted to a few diseases and specific equipment. Here, we present an end-to-end deep learning framework named AIEchoDx that differentiates four common cardiovascular diseases (Atrial Septal Defect, Dilated Cardiomyopathy, Hypertrophic Cardiomyopathy, prior Myocardial Infarction) from normal subjects with performance comparable to that of consensus of three senior cardiologists in AUCs (99.50% vs 99.26%, 98.75% vs 92.75%, 99.57% vs 97.21%, 98.52% vs 84.20%, and 98.70% vs 89.41%), respectively. Meanwhile, AIEchoDx accurately recognizes critical lesion regions of interest along with each disease by visualizing the decision-making process. Furthermore, our analysis indicates that heterogeneous diseases, like dilated cardiomyopathy, could be classified into two phenogroups with distinct clinical characteristics. Finally, AIEchoDx performs efficiently as an anomaly detection tool when applying handheld device-produced videos. Together, AIEchoDx provides a potential diagnostic assistant tool in either cart-based echocardiography equipment or handheld echocardiography device for primary and point-of-care medical personnel with high diagnostic performance, and the application of lesion region identification and heterogeneous disease phenogrouping, which may broaden the application of artificial intelligence in echocardiography. [ABSTRACT FROM AUTHOR]

Although ship detection methods based on deep learning have achieved remarkable progress, the design of the universal ship detection (USD) system is rarely studied. The main challenge of USD lies in the notorious domain bias and shift problem across multiple domains. This article implements USD based on domain-invariant representations to alleviate this issue. Specifically, the proposed method integrates a multilevel domain classification network (MDCN) and a domain-centric cut-paste module (DCM). First, the backbone network is facilitated to learn domain-independent image features from multiple domains through MDCN, thereby reducing the disturbance of domain-specific features to universal detector. Furthermore, the proposed method combines the domain-related synthetic samples generated by DCM to provide MDCN with strong supervision information, which further motivates the network to be more attentive to the domain-invariant representations at the instance level. Finally, we conduct experiments on multiple ship datasets in the synthetic aperture radar (SAR) and optical domains to verify the effectiveness of the method. The results show that the proposed method outperforms baseline by around 2.95% average precision (AP50), which achieves an effective USD system by complementing the information between domain-invariant representations. [ABSTRACT FROM AUTHOR]

Deep learning analysis of images and text unfolds new horizons in medicine. However, analysis of transcriptomic data, the cause of biological and pathological changes, is hampered by structural complexity distinctive from images and text. Here we conduct unsupervised training on more than 20,000 human normal and tumor transcriptomic data and show that the resulting Deep-Autoencoder, DeepT2Vec, has successfully extracted informative features and embedded transcriptomes into 30-dimensional Transcriptomic Feature Vectors (TFVs). We demonstrate that the TFVs could recapitulate expression patterns and be used to track tissue origins. Trained on these extracted features only, a supervised classifier, DeepC, can effectively distinguish tumors from normal samples with an accuracy of 90% for Pan-Cancer and reach an average 94% for specific cancers. Training on a connected network, the accuracy is further increased to 96% for Pan-Cancer. Together, our study shows that deep learning with autoencoder is suitable for transcriptomic analysis, and DeepT2Vec could be successfully applied to distinguish cancers, normal tissues, and other potential traits with limited samples. [ABSTRACT FROM AUTHOR]

Recent advances in deep learning for medical image segmentation demonstrate expert-level accuracy. However, application of these models in clinically realistic environments can result in poor generalization and decreased accuracy, mainly due to the domain shift across different hospitals, scanner vendors, imaging protocols, and patient populations etc. Common transfer learning and domain adaptation techniques are proposed to address this bottleneck. However, these solutions require data (and annotations) from the target domain to retrain the model, and is therefore restrictive in practice for widespread model deployment. Ideally, we wish to have a trained (locked) model that can work uniformly well across unseen domains without further training. In this paper, we propose a deep stacked transformation approach for domain generalization. Specifically, a series of ${n}$ stacked transformations are applied to each image during network training. The underlying assumption is that the “expected” domain shift for a specific medical imaging modality could be simulated by applying extensive data augmentation on a single source domain, and consequently, a deep model trained on the augmented “big” data (BigAug) could generalize well on unseen domains. We exploit four surprisingly effective, but previously understudied, image-based characteristics for data augmentation to overcome the domain generalization problem. We train and evaluate the BigAug model (with ${n}={9}$ transformations) on three different 3D segmentation tasks (prostate gland, left atrial, left ventricle) covering two medical imaging modalities (MRI and ultrasound) involving eight publicly available challenge datasets. The results show that when training on relatively small dataset (n = 10~32 volumes, depending on the size of the available datasets) from a single source domain: (i) BigAug models degrade an average of 11%(Dice score change) from source to unseen domain, substantially better than conventional augmentation (degrading 39%) and CycleGAN-based domain adaptation method (degrading 25%), (ii) BigAug is better than “shallower” stacked transforms (i.e. those with fewer transforms) on unseen domains and demonstrates modest improvement to conventional augmentation on the source domain, (iii) after training with BigAug on one source domain, performance on an unseen domain is similar to training a model from scratch on that domain when using the same number of training samples. When training on large datasets (n = 465 volumes) with BigAug, (iv) application to unseen domains reaches the performance of state-of-the-art fully supervised models that are trained and tested on their source domains. These findings establish a strong benchmark for the study of domain generalization in medical imaging, and can be generalized to the design of highly robust deep segmentation models for clinical deployment. [ABSTRACT FROM AUTHOR]

Coastal ship tracking is used in many applications, such as autonomous navigation, maritime rescue, and environmental monitoring. Many general object-tracking methods based on deep learning have been explored for ship tracking, but they often fail to accurately track ships in challenging scenarios, such as occlusion, scale variation, and motion blur. We propose a memory-guided perception network (MGPN) to address these issues. MGPN has two main innovative improvements. The dynamic memory mechanism (DMM) in the proposed method stores past features of the tracked target to enhance the model's feature fusion capability in the temporal dimension. Meanwhile, the hierarchical context-aware module (HCAM) enables the interaction of different scales, global and local information, to address the scale discrepancy of targets and improve the feature fusion capability in the spatial dimension. These innovations enhance the robustness of tracking and reduce inaccuracies in the bounding boxes. We conducted an in-depth ablation study to demonstrate the effectiveness of DMM and HCAM. Finally, influenced by the above two points, MGPN has achieved state-of-the-art performance on a large offshore ship tracking dataset, which contains challenging scenarios such as complex backgrounds, ship occlusion, and varying scales. [ABSTRACT FROM AUTHOR]

Indocyanine green (ICG) has been used in clinical practice for more than 40 years and its safety and preferential accumulation in tumors has been reported for various tumor types, including colon cancer. However, reports on clinical assessments of ICG-based molecular endoscopy imaging for precancerous lesions are scarce. We determined visualization ability of ICG fluorescence endoscopy in colitis-associated colon cancer using 30 lesions from an azoxymethane/dextran sulfate sodium (AOM/DSS) mouse model and 16 colon cancer patient tissue-samples. With a total of 60 images (optical, fluorescence) obtained during endoscopy observation of mouse colon cancer, we used deep learning network to predict four classes (Normal, Dysplasia, Adenoma, and Carcinoma) of colorectal cancer development. ICG could detect 100% of carcinoma, 90% of adenoma, and 57% of dysplasia, with little background signal at 30 min after injection via real-time fluorescence endoscopy. Correlation analysis with immunohistochemistry revealed a positive correlation of ICG with inducible nitric oxide synthase (iNOS; r > 0.5). Increased expression of iNOS resulted in increased levels of cellular nitric oxide in cancer cells compared to that in normal cells, which was related to the inhibition of drug efflux via the ABCB1 transporter down-regulation resulting in delayed retention of intracellular ICG. With artificial intelligence training, the accuracy of image classification into four classes using data sets, such as fluorescence, optical, and fluorescence/optical images was assessed. Fluorescence images obtained the highest accuracy (AUC of 0.8125) than optical and fluorescence/optical images (AUC of 0.75 and 0.6667, respectively). These findings highlight the clinical feasibility of ICG as a detector of precancerous lesions in real-time fluorescence endoscopy with artificial intelligence training and suggest that the mechanism of ICG retention in cancer cells is related to intracellular nitric oxide concentration. [ABSTRACT FROM AUTHOR]

Although the existing deep-learning-based hyperspectral image (HSI) denoising methods have achieved tremendous success, recovering high-quality HSIs in complex scenes that contain mixed noise is still challenging. Besides, these methods have not fully explored the local and global spatial–spectral information of HSIs. To address the above issues, a novel HSI mixed noise removal network called subspace projection attention and residual channel attention network (SPARCA-Net) is proposed. Specifically, we propose an orthogonal subspace projection attention (OSPA) module to adaptively learn to generate bases of the signal subspace and project the input into such space to remove noise. By leveraging the local and global spatial relations, OSPA is able to reconstruct the local structure of the feature maps more precisely. We further propose a residual channel attention (RCA) module to emphasize the interdependence between feature maps and exploit the global channel correlation of them, which could enhance the channel-wise adaptive learning. In addition, multiscale joint spatial–spectral input and residual learning strategies are employed to capture multiscale spatial–spectral features and reduce the degradation problem, respectively. Synthetic and real HSI data experiments demonstrated that the proposed HSI denoising network outperforms many of the advanced methods in both quantitative and qualitative assessments. [ABSTRACT FROM AUTHOR]

A better understanding of temporal relationships between chest CT and labs may provide a reference for disease severity over the disease course. Generalized curves of lung opacity volume and density over time can be used as standardized references from well before symptoms develop to over a month after recovery, when residual lung opacities remain. 739 patients with COVID-19 underwent CT and RT-PCR in an outbreak setting between January 21st and April 12th, 2020. 29 of 739 patients had serial exams (121 CTs and 279 laboratory measurements) over 50 ± 16 days, with an average of 4.2 sequential CTs each. Sequential volumes of total lung, overall opacity and opacity subtypes (ground glass opacity [GGO] and consolidation) were extracted using deep learning and manual segmentation. Generalized temporal curves of CT and laboratory measurements were correlated. Lung opacities appeared 3.4 ± 2.2 days prior to symptom onset. Opacity peaked 1 day after symptom onset. GGO onset was earlier and resolved later than consolidation. Lactate dehydrogenase, and C-reactive protein peaked earlier than procalcitonin and leukopenia. The temporal relationships of quantitative CT features and clinical labs have distinctive patterns and peaks in relation to symptom onset, which may inform early clinical course in patients with mild COVID-19 pneumonia, or may shed light upon chronic lung effects or mechanisms of medical countermeasures in clinical trials. [ABSTRACT FROM AUTHOR]

The blockchain whitepaper contains detailed technical and business information, so its analysis is important for blockchain text mining. Previous works focus on analyze homogeneous objects and relations. The main problem, however, is these works do not take into account the heterogeneity of information. This paper presents a new methodology for whitepapers analysis by designing heterogeneous graph neural network, named S-HGNN. In detail, this paper first builds a Heterogeneous Information Network (HIN) using heterogeneous objects and relationships extracted from the whitepaper to obtain similarity measures, then uses Graph Convolutional Network (GCN) and Graph Attention Network (GAT) to integrate both structural information and internal semantic into the whitepaper embedding. Compared with the previous models, this model improves 0.96% ∼ 33.34% in terms of F1-score for classification task, and 4.94% ∼ 14.14% in terms of purity for clustering task, and gets stable results on different tasks. The results show the effectiveness and robustness of this model for whitepapers analysis. • We propose a novel automatic blockchain whitepapers analysis framework which integrates both popular data mining and deep learning techniques for intelligent analysis of whitepapers about blockchain products. • This paper explores a heterogeneous Graph Neural Network based on GCN and GAT, and applies the learned feature of whitepapers into clustering and classification tasks. [ABSTRACT FROM AUTHOR]

We developed and validated a deep-learning algorithm for polyp detection. We used a YOLOv2 to develop the algorithm for automatic polyp detection on 8,075 images (503 polyps). We validated the algorithm using three datasets: A: 1,338 images with 1,349 polyps; B: an open, public CVC-clinic database with 612 polyp images; and C: 7 colonoscopy videos with 26 polyps. To reduce the number of false positives in the video analysis, median filtering was applied. We tested the algorithm performance using 15 unaltered colonoscopy videos (dataset D). For datasets A and B, the per-image polyp detection sensitivity was 96.7% and 90.2%, respectively. For video study (dataset C), the per-image polyp detection sensitivity was 87.7%. False positive rates were 12.5% without a median filter and 6.3% with a median filter with a window size of 13. For dataset D, the sensitivity and false positive rate were 89.3% and 8.3%, respectively. The algorithm detected all 38 polyps that the endoscopists detected and 7 additional polyps. The operation speed was 67.16 frames per second. The automatic polyp detection algorithm exhibited good performance, as evidenced by the high detection sensitivity and rapid processing. Our algorithm may help endoscopists improve polyp detection. [ABSTRACT FROM AUTHOR]

We aimed to develop a computer-aided diagnostic system (CAD) for predicting colorectal polyp histology using deep-learning technology and to validate its performance. Near-focus narrow-band imaging (NBI) pictures of colorectal polyps were retrieved from the database of our institution. Of these, 12480 image patches of 624 polyps were used as a training set to develop the CAD. The CAD performance was validated with two test datasets of 545 polyps. Polyps were classified into three histological groups: serrated polyp (SP), benign adenoma (BA)/mucosal or superficial submucosal cancer (MSMC), and deep submucosal cancer (DSMC). The overall kappa value measuring the agreement between the true polyp histology and the expected histology by the CAD was 0.614–0.642, which was higher than that of trainees (n = 6, endoscopists with experience of 100 NBI colonoscopies in <6 months; 0.368–0.401) and almost comparable with that of the experts (n = 3, endoscopists with experience of 2,500 NBI colonoscopies in ≥5 years) (0.649–0.735). The areas under the receiver operating curves for CAD were 0.93–0.95, 0.86–0.89, and 0.89–0.91 for SP, BA/MSMC, and DSMC, respectively. The overall diagnostic accuracy of the CAD was 81.3–82.4%, which was significantly higher than that of the trainees (63.8–71.8%, P < 0.01) and comparable with that of experts (82.4–87.3%). The kappa value and diagnostic accuracies of the trainees improved with CAD assistance: that is, the kappa value increased from 0.368 to 0.655, and the overall diagnostic accuracy increased from 63.8–71.8% to 82.7–84.2%. CAD using a deep-learning model can accurately assess polyp histology and may facilitate the diagnosis of colorectal polyps by endoscopists. [ABSTRACT FROM AUTHOR]

Background: Neuropsychological tests (NPTs) are important tools for informing diagnoses of cognitive impairment (CI). However, interpreting NPTs requires specialists and is thus time-consuming. To streamline the application of NPTs in clinical settings, we developed and evaluated the accuracy of a machine learning algorithm using multi-center NPT data.Methods: Multi-center data were obtained from 14,926 formal neuropsychological assessments (Seoul Neuropsychological Screening Battery), which were classified into normal cognition (NC), mild cognitive impairment (MCI) and Alzheimer's disease dementia (ADD). We trained a machine learning model with artificial neural network algorithm using TensorFlow (https://www.tensorflow.org) to distinguish cognitive state with the 46-variable data and measured prediction accuracies from 10 randomly selected datasets. The features of the NPT were listed in order of their contribution to the outcome using Recursive Feature Elimination.Results: The ten times mean accuracies of identifying CI (MCI and ADD) achieved by 96.66 ± 0.52% of the balanced dataset and 97.23 ± 0.32% of the clinic-based dataset, and the accuracies for predicting cognitive states (NC, MCI or ADD) were 95.49 ± 0.53 and 96.34 ± 1.03%. The sensitivity to the detection CI and MCI in the balanced dataset were 96.0 and 96.0%, and the specificity were 96.8 and 97.4%, respectively. The 'time orientation' and '3-word recall' score of MMSE were highly ranked features in predicting CI and cognitive state. The twelve features reduced from 46 variable of NPTs with age and education had contributed to more than 90% accuracy in predicting cognitive impairment.Conclusions: The machine learning algorithm for NPTs has suggested potential use as a reference in differentiating cognitive impairment in the clinical setting. [ABSTRACT FROM AUTHOR]