Management of cN0 papillary thyroid microcarcinoma patients according to risk-scoring model for central lymph node metastasis and predictors of recurrence.

BACKGROUND: The role of routine prophylactic central neck dissection (pCND) in clinically lymph node-negative (cN0) papillary thyroid microcarcinoma (PTMC) patients remains controversial. This retrospective study aimed to identify the clinical and pathologic factors of central lymph node metastasis (CLNM) and recurrence in PTMC patients. METHODS: A total of 371 cN0 PTMC patients from two hospitals were retrospectively analyzed. All patients underwent thyroidectomy plus pCND between January 2010 and January 2018. Clinicopathological features were collected, univariate and multivariate analyses were performed to determine the risk factors of CLNM. A scoring model was constructed on the basis of the results of independent risk factors of CLNM. The Cox proportional hazards model was used to analyze the risk factors of recurrence. RESULTS: CLNM occurred in 123 (33.2%) patients. Multivariate analysis showed male, tumor size > 0.75 cm, multifocality, extrathyroidal extension (ETE) and tumor in the middle/lower pole were independent risk predictors of CLNM (P < 0.05). A seven-point risk-scoring model was established to predict the stratified CLNM in cN0 PTMC patients. Multivariate Cox regression model showed ETE, vascular invasion and CLNM were independent risk predictors of recurrence (P < 0.05). CONCLUSION: Our study suggested that routine pCND should be performed for cN0 PTMC patients with score ≥ 3 according to the risk-scoring model. Moreover, patients with risk factors of recurrence should consider more complete treatment and more frequent follow-up.

[Effect of serum IgA/C3 and glomerular C3 staining on clinical prognosis in patients with IgA nephropathy].

Objective: To investigate the effects of serum immunoglobulin A/complement factor 3 (IgA/C3) ratio and glomerular C3 staining on clinical prognosis in patients with IgA nephropathy. Methods: From January 1st, 2007 to December 30th, 2016, a total of 519 patients with biopsy-proven IgA nephropathy (IgAN) in West China Hospital were retrospectively reviewed and divided into four groups based on serum IgA/C3 ratio and glomerular C3 staining: group A with IgA/C3 ratio ≥3.046 (median) and glomerular C3 staining ≥2 (n=151), group B with IgA/C3 ratio ≥3.046 and glomerular C3 staining<2 (n=109), group C with IgA/C3 ratio<3.046 and glomerular C3 staining ≥2 (n=119), and group D with IgA/C3 ratio<3.046 and glomerular C3 staining<2 (n=140). Clinical data, pathological characteristics and the primary endpoint [≥ 50% decline in estimated glomerular filtration rate (eGFR) and/or end-stage renal disease (ESRD)]were collected. Clinical prognosis and relevant risk factors were analyzed among the four groups. Results: Totally, 519 patients (298 males, 57.4%) with an average age of (33.6±10.9) years were recruited and followed up for (43.4±21.6) months. The rate of complete remission plus partial remission was 74.2% (112/151), 74.3% (81/109), 72.3% (86/119), 81.4% (114/140) in group A, B, C, D, respectively. Meanwhile, The rate of ESRD was highest in group A (14.6% vs 9.2%, 13.4%, 8.6%). Renal outcome (patients reached the endpoint) was worse in group A and C compared with group B and D (15.2%, 16.0 vs 8.3%, 7.9%). Moreover, 80-month renal survival rate was significantly worse in group A (84.8%) than that in group B and D (91.7% and 92.1%), but no statistical significant difference was found between group A and B (P(AB)=0.085; P(AD)=0.028). There was no significant difference of renal survival rate between group A and C (84.8% vs 84.0%, P=0.896). Multivariate Cox model showed that hypertension (HR=2.753, 95%CI: 1.452-5.217, P=0.002), serum creatinine (HR=1.011, 95%CI: 1.008-1.014, P<0.001), and tubular atrophy/interstitial fibrosis (T1/T2) (HR=6.595, 95%CI: 3.107-13.999, P<0.001) were independent predictors of poor renal survival. Conclusion: Serum IgA/C3 ratio and glomerular C3 staining are predictors of renal clinical prognosis in patients with IgA nephropathy.

Biphasic hierarchical extracellular matrix scaffold for osteochondral defect regeneration.

OBJECTIVE: To investigate the effect of decellularized osteochondral extracellular matrix (ECM) scaffold for osteochondral defect regeneration. DESIGN: We compared the histological features and microstructure of degenerated cartilage to normal articular cartilage. We also generated and evaluated osteochondral ECM scaffolds through decellularization technology. Then scaffolds were implanted to osteochondral defect in rabbit model. After 12 weeks surgery, regeneration tissues were analyzed by histology, immunohistochemistry evaluation. And possible mechanisms of angiogenesis and cell migration were explored. RESULTS: We demonstrated decreased cell numbers, formation of fibrous cartilage, lost microstructure and worse permeability in degenerated cartilage compared to normal cartilage. We also generated an osteochondral ECM scaffold with a hierarchical structure that exhibited low immunogenicity, high bioactivity, and well biocompatibility. We found that the ECM scaffold promoted tissue regeneration in osteochondral defects, which was dependent on the scaffold constituents and stratified three-dimensional microstructure as well as on its ability to inhibit angiogenesis and stimulate cell migration. CONCLUSIONS: Our findings demonstrated that the biphasic hierarchical ECM scaffold represents a novel and effective biomaterial that can be used in the treatment of osteochondral defect.

[The effects of microRNA-7 on proliferation and invasion of hepatocellular carcinoma HepG2 cells].

Objective: To investigate the effects of overexpression of microRNA-7 (miR-7) on the proliferation and invasion of HepG2 cells and the underlying mechanism in vitro. Methods: The relative expression levels of miR-7 and Raf1 in hepatocellular carcinoma (HCC) tissues and adjacent normal tissues (ANT) were detected by quantitative real time-PCR (qRT-PCR). The relationship between the expression of miR-7 and the characteristics of HCC patients was analyzed. Cells were divided into blank control group, negative control (NC) group and miR-7 mimics transfected group, miR-7 mimics and NC were transfected into HepG2 cells by Lipofectamine™2000. The relative expression of miR-7 was detected by qRT-PCR. The proliferation ability of HepG2 cells was detected by 3-(4, 5-dimethyl-2-thiazolyl)-2, 5-diphenyl-2-H-tetrazolium bromide (MTT) assay. The invasion of HepG2 cells was detected by Transwell assay. The target genes of miR-7 were predicted by TargetScan and the binding effect of miR-7 on the 3'UTR of Raf1 was verified by dual luciferase reporter assay.The expressions of Raf1 protein in hepatocellular carcinoma tissues, normal tissues and miR-7 mimics transfected HepG2 cells was detected by Western blot. The correlation of the levels of miR-7 and Raf1 mRNA was determined by Pearson correlation analysis. Results: The relative expression level of miR-7 in HCC was 0.49±0.02, significantly lower than in ANT (1.21±0.05, P<0.01). The level of miR-7 was significantly correlated the tumor volume, metastasis and prognosis of HCC patients (P<0.05). The relative expression level of miR-7 in miR-7 mimics transfected HepG2 group was 12.67±0.40, significantly higher than that in blank group (P<0.01). Compared with the blank group, the A value and invasion ability of miR-7 mimics transfected group were significantly down-regulated at 48 hours and 72 hours after transfection (P<0.01). Compared with miR-7 NC group, the luciferase activity of wild-type Raf1 reporter gene in miR-7 mimics transfected group was significantly reduced (P<0.01). The relative expression of Raf1 protein in HCC was 3.15±0.10, significant higher than in ANT (0.53±0.03, P<0.01). The relative expression of Raf1 protein in miR-7 mimics transfected group was 0.24±0.01, significantly lower than in miR-7 NC group (0.98±0.02, P<0.01). Furthermore, an negative correlation was observed between the levels of miR-7 and Raf1 in HCC tissues (P<0.05). Conclusions: The expression of miR-7 in HCC is significantly decreased and inversely correlated with poor survival of HCC patients. Overexpression of miR-7 can inhibit the proliferation and invasion ability of hepatocellular carcinoma cells HepG2 by downregulating Raf1 in vitro.

A phase II study of a human anti-PDGFRα monoclonal antibody (olaratumab, IMC-3G3) in previously treated patients with metastatic gastrointestinal stromal tumors.

Background: This study evaluated tumor response to olaratumab (an anti-PDGFRα monoclonal antibody) in previously treated patients with metastatic gastrointestinal stromal tumor (GIST) with or without PDGFRα mutations (cohorts 1 and 2, respectively). Patients and methods: Patients received olaratumab 20 mg/kg intravenously every 14 days until disease progression, death, or intolerable toxicity occurred. Outcome measures were 12-week tumor response, progression-free survival (PFS), overall survival (OS), and safety. Results: Of 30 patients enrolled, 21 patients received ≥1 dose of olaratumab. In the evaluable population (cohort 1, n = 6; cohort 2, n = 14), no complete response (CR) or partial response (PR) was observed. Stable disease (SD) was observed in 3 patients (50.0%) in cohort 1 and 2 patients (14.3%) in cohort 2. Progressive disease (PD) was observed in 3 patients (50.0%) in cohort 1 and 12 patients (85.7%) in cohort 2. The 12-week clinical benefit rate (CR + PR + SD) (90% CI) was 50.0% (15.3-84.7%) in cohort 1 and 14.3% (2.6-38.5%) in cohort 2. SD lasted beyond 12 weeks in 5 patients (cohort 1, n = 3; cohort 2, n = 2). Median PFS (90% CI) was 32.1 (5.0-35.9) weeks in cohort 1 and 6.1 (5.7-6.3) weeks in cohort 2. Median OS was not reached in cohort 1 and was 24.9 (14.4-49.1) weeks in cohort 2. All patients in cohort 1 and 9 (64.3%) in cohort 2 experienced an olaratumab-related adverse event (AE), most commonly fatigue (38.1%), nausea (19.0%), and peripheral edema (14.3%). Two grade ≥3 olaratumab-related events were reported (cohort 1, syncope; cohort 2, hypertension). Conclusions: Olaratumab had an acceptable AE profile in patients with GIST. While there was no apparent effect on PFS in patients without PDGFRα mutations, patients with PDGFRα-mutant GIST (all with D842V mutations) treated with olaratumab had longer disease control compared with historical data for this genotype. ClinicalTrials.gov Identifier: NCT01316263.

Clinical Significance of Myeloid-Derived Suppressor Cells in Human Immunodeficiency Virus-1/ Hepatitis C Virus-coinfected Patients.

Myeloid-derived suppressor cells (MDSCs) are known to accumulate during chronic viral infection, including human immunodeficiency virus-1 (HIV-1) and hepatitis C virus (HCV) infection, and play a critical role in suppressing immune responses. However, the role of MDSCs in HIV/HCV coinfection is unclear. Here, we observed a dramatic increase in monocytic MDSCs (M-MDSCs) level in the peripheral blood of HIV/HCV-coinfected patients compared to that of healthy controls; the level of M-MDSCs proportion in coinfection was not higher than that in HIV or HCV monoinfection. Interestingly, we found the M-MDSCs level in coinfected patients correlated well with CD4(+) T cell loss (r = -0.5680; P = 0.0058), HIV-1 load (r = 0.6011; P = 0.0031), HCV load (r = 0.6288; P = 0.0017) and activated CD38(+) T cells (r = 0.5139; P = 0.0144). Initiation of highly active antiretroviral therapy considerably reduced both M-MDSCs and CD8(+) CD38(+) -activated T cell proportion in coinfected patients, and they showed a parallel course of decline. Thus, our results suggest that HIV-1 infection and high chronic immune activation may contribute to the expansion of M-MDSCs and accelerate the disease progression in HIV/HCV-coinfected patients.

The rate of co-infection for piglet diarrhea viruses in China and the genetic characterization of porcine epidemic diarrhea virus and porcine kobuvirus.

Piglet diarrhea epidemics result in major economic losses for the swine industry. Four viruses are closely linked to porcine diarrhea: porcine kobuvirus (PKV), porcine epidemic diarrhea virus (PEDV), porcine transmissible gastroenteritis virus (TGEV), and porcine rotavirus (PRoV). We have conducted an epidemiology study to determine the frequency of infection and co-infection with these viruses in China, and characterized the genetic variation of the isolated PEDV and PKV strains. Stool and intestinal samples (n = 314) were collected from piglets with diarrhea in China from years 2012 to 2014. RT-PCR was used to detect PKV, PEDV, TGEV, and PRoV. Phylogenetic relationships between reference strains and the isolated PEDV and PKV strains were determined based on the M and 3D gene sequence. The rates of infection with PKV, PEDV, TGEV and PRoV were 29.9%, 24.2%, 1.91%, and 0.31%, respectively. Co-infections with PKV and the other three viruses were very common. Co-infection of PKV and PEDV was detected in 15.0% (47/314) of the samples. Phylogenetic analysis of the PKV 3D gene indicated that there were some phylogenetic differences in the PKV strains across regions within China. However, according to the PEDV M gene, strains clustered into three groups and the primary group was distinct from the vaccine strain CV777. This study provides insights in to the prevalence of diarrhea viruses and their prevention and control in China.

Molecular characterization, expression, and functional analysis of NOD1 in Qingyuan partridge chicken.

Nucleotide-binding oligomerization domain-containing protein-1 (NOD1) is a cytoplasmic pattern recognition receptor (PRR) and a key member of the NOD-like receptor (NLR) family. It has been reported that NLRs recognize a variety of microbial infections to induce the host innate immune response via modulation of NF-κB signaling. However, no reports on chicken NOD1 have been reported to date. In the current study, the full-length cDNA sequence of NOD1 was cloned. The complete open reading frame of NOD1 contains 2856 bp and encodes a 951 amino acid protein. Structurally, it is comprised of one caspase recruitment domain at the N-terminus, seven leucine-rich repeat regions at the C-terminus, and one NACHT domain between the N and C-termini. Phylogenetic analyses showed that chicken NOD1 clusters with duck and turkey. Furthermore, tissue-specific expression analyses of chicken NOD1 were performed using quantitative reverse transcription-PCR. NOD1 is widely distributed in various tissues, with the highest expression observed in testes. Finally, induced expression of chNOD1 and its associated adaptor molecule receptor-interacting protein 2, as well as the effector molecule NF-κB, was observed following S. enterica serovar Enteritidis infection. These findings highlight the important role of chicken NOD1 in response to pathogenic invasion. The present study is the first report of the cloning, expression, and functional analysis of chicken NOD1 and provides the foundation for future research on the structure and function of chicken NOD1.

OA10 is a novel p38alpha mitogen-activated protein kinase inhibitor that suppresses osteoclast differentiation and bone resorption.

In search of anti-bone resorbing agents for the potential treatment of osteoporosis, we synthesized a novel compound Tert-butyl 4-(3-[1H-indole-2-carboxamido]benzoyl)piperazine-1-carboxylate (OA10) and found that OA10 is capable of inhibiting RANKL-mediated osteoclast formation and osteoclastic bone resorption in a dose-dependent manner. This biological effect is further supported by the fact that OA10 suppressed osteoclastic-specific gene expression, including tartrate-resistant acid phosphatase, cathepsin K receptor, and calcitonin receptor. Further molecular mechanism investigation revealed OA10 inhibited p38 phosphorylation, suppressed c-fos and NFATc1 expression without affecting NF-κB or JNK signaling pathways. Taken together, this study suggested that OA10 can inhibit osteoclastogenesis by suppressing p38-c-Fos-NFATc1 cascade. OA10 may be developed as a therapeutic drug for osteoclast-related osteolytic diseases.

Identical subchondral bone microarchitecture pattern with increased bone resorption in rheumatoid arthritis as compared to osteoarthritis.

OBJECTIVES: To analyze the differences in microarchitecture and bone remodeling of subchondral bone in femoral heads from patients with rheumatoid arthritis (RA) and osteoarthritis (OA). DESIGNS: Peri-articular bone samples, including subchondral trabecular bone (STB) and deeper trabecular bone (DTB) were extracted from the load-bearing region of femoral heads from 20 patients with RA and 40 patients with OA during hip replacement surgery. Micro-CT, histomorphometry and backscatter scanning electron microscopy (BSEM) were performed to assess microarchitecture and bone histology parameters. RESULTS: In both RA and OA, STB showed more sclerotic microarchitecture and more active bone remodeling, compared to DTB. RA and OA showed similar microarchitecture characteristics in both STB and DTB, despite STB in RA exhibiting higher bone resorption. In addition, there was no difference in the frequency of bone cysts in STB between RA and OA. In STB, the trabecular bone surrounding subchondral bone cysts (Cys-Tb) was more sclerotic than the trabecular bone found distant to cysts (Peri-Tb), with a higher level of bone remodeling. Both Cys-Tb region and Peri-Tb region were detected to have similar microarchitectural and bone remodeling characteristics in RA and OA. CONCLUSIONS: Apart from higher bone resorption in the general subchondral bone of RA samples, the peri-articular bone exhibited similar microarchitectural and bone remodeling characteristics in RA and OA.

Coffee, tea, and the risk of hip fracture: a meta-analysis.

UNLABELLED: The present meta-analysis shows no clear association between coffee consumption and the risk of hip fractures. There was a nonlinear association between tea consumption and the risk of hip fracture. Compared to no tea consumption, drinking 1-4 cups of tea daily was associated with a lower risk of hip fracture. INTRODUCTION: Prospective cohort and case-control studies have suggested that coffee and tea consumption may be associated with the risk of hip fracture; the results have, however, been inconsistent. We conducted a meta-analysis to assess the association between coffee and tea consumption and the risk of hip fracture. METHODS: We performed systematic searches using MEDLINE, EMBASE, and OVID until February 20, 2013, without limits of language or publication year. Relative risks (RRs) with 95% confidence intervals (CI) were derived using random-effects models throughout all analyses. We conducted categorical, dose-response, heterogeneity, publication bias, and subgroup analyses. RESULTS: Our study was based on 195,992 individuals with 9,958 cases of hip fractures from 14 studies, including six cohort and eight case-control studies. The pooled RRs of hip fractures for the highest vs. the lowest categories of coffee and tea consumption were 0.94 (95% CI 0.71-1.17) and 0.84 (95% CI 0.66-1.02), respectively. For the dose-response analysis, we found evidence of a nonlinear association between tea consumption and the risk of hip fracture (p(nonlinearity) < 0.01). Compared to no tea consumption, 1-4 cups of tea per day may reduce the risk of hip fracture by 28% (0.72; 95% CI 0.56-0.88 for 1-2 cups/day), 37% (0.63; 95% CI 0.32-0.94 for 2-3 cups/day), and 21% (0.79; 95% CI 0.62-0.96 for 3-4 cups/day). CONCLUSIONS: We found no significant association between coffee consumption and the risk of hip fracture. A nonlinear association emerged between tea consumption and the risk of hip fracture; individuals drinking 1-4 cups of tea per day exhibited a lower risk of hip fractures than those who drank no tea. The association between 5 daily cups of tea, or more, and hip fracture risk should be investigated.

Isolation of a goose parvovirus from swan and its molecular characteristics.

Goose parvovirus (GPV) causes high mortality and morbidity in goslings and Muscovy ducklings. In this study, a GPV was isolated from a 20-day old swan in Shanghai, China. Complete genome of the swan isolate contained 5,050 nt and showed the highest homology with Taiwanese GPV isolates from 1982. In comparison with the Chinese mainland GPV isolates reported previously, the swan isolate shows two deletions, particularly at positions 67-80 and 334-347 in inverted terminal repeats (ITRs). These findings suggest that the swan could serve as a potential host for GPV and provide insights into molecular characteristics and etiology of GPV.

Differential protein analysis of chicken skin infected with Marek΄s disease virus.

The skin and feather follicle epithelia of birds infected with Marek's disease virus (MDV) are the sites of infectious virus particle formation and shedding. However, the host responses and protein networks involved in the production of virus particles in the skin of MDV-infected chickens are poorly understood. This current study aimed to analyze the differential protein expression patterns in skin between MDV-infected and uninfected specific pathogen-free (SPF) chickens 28 days post infection (dpi) by combining two-dimensional electrophoresis (2-DE) and mass spectrometry (MS) analyses. Through 2-DE analysis, our results revealed 23 proteins whose expression changed significantly following infection, of which 16 proteins were confirmed by MS. The identified proteins were functionally classified into 5 groups: immune-related, cell regulatory, cytoskeletal, metabolism-related and transport proteins. A single protein, beta 2-microglobulin, was further confirmed by real-time quantitative PCR. Beta 2-microglobulin expression was significantly increased in the infected group 28 dpi. This indicates that beta 2-microglobulin might play very important roles in the viral evasion from host immune response.

Toward Explainable Multiparty Learning: A Contrastive Knowledge Sharing Framework.

Multiparty learning provides solutions for training joint models with decentralized data under legal and practical constraints. However, traditional multiparty learning approaches are confronted with obstacles, such as system heterogeneity, statistical heterogeneity, and incentive design. Determining how to deal with these challenges and further improve the efficiency and performance of multiparty learning has become an urgent problem to be solved. In this article, we propose a novel contrastive multiparty learning framework for knowledge refinement and sharing with an accountable incentive mechanism. Since the existing parameter averaging method is contradictory to the learning paradigm of neural networks, we simulate the process of human cognition and communication and analogize multiparty learning as a many-to-one knowledge-sharing problem. The approach is capable of integrating the acquired explicit knowledge of each client in a transparent manner without privacy disclosure, and it reduces the dependence on data distribution and communication environments. The proposed scheme achieves significant improvement in model performance in a variety of scenarios, as we demonstrated through experiments on several real-world datasets.

Self-Paced Co-Training of Graph Neural Networks for Semi-Supervised Node Classification.

Graph neural networks (GNNs) have demonstrated great success in many graph data-based applications. The impressive behavior of GNNs typically relies on the availability of a sufficient amount of labeled data for model training. However, in practice, obtaining a large number of annotations is prohibitively labor-intensive and even impossible. Co-training is a popular semi-supervised learning (SSL) paradigm, which trains multiple models based on a common training set while augmenting the limited amount of labeled data used for training each model via the pseudolabeled data generated from the prediction results of other models. Most of the existing co-training works do not control the quality of pseudolabeled data when using them. Therefore, the inaccurate pseudolabels generated by immature models in the early stage of the training process are likely to cause noticeable errors when they are used for augmenting the training data for other models. To address this issue, we propose a self-paced co-training for the GNN (SPC-GNN) framework for semi-supervised node classification. This framework trains multiple GNNs with the same or different structures on different representations of the same training data. Each GNN carries out SSL by using both the originally available labeled data and the augmented pseudolabeled data generated from other GNNs. To control the quality of pseudolabels, a self-paced label augmentation strategy is designed to make the pseudolabels generated at a higher confidence level to be utilized earlier during training such that the negative impact of inaccurate pseudolabels on training data augmentation, and accordingly, the subsequent training process can be mitigated. Finally, each of the trained GNN is evaluated on a validation set, and the best-performing one is chosen as the output. To improve the training effectiveness of the framework, we devise a pretraining followed by a two-step optimization scheme to train GNNs. Experimental results on the node classification task demonstrate that the proposed framework achieves significant improvement over the state-of-the-art SSL methods.

DANet: Semi-supervised differentiated auxiliaries guided network for video action recognition.

Video Action Recognition (ViAR) aims to identify the category of the human action observed in a given video. With the advent of Deep Learning (DL) techniques, noticeable performance breakthroughs have been achieved in this study. However, the success of most existing DL-based ViAR methods heavily relies on the existence of a large amount of annotated data, i.e., videos with corresponding action categories. In practice, obtaining such a desired number of annotations is often difficult due to expensive labeling costs, which may lead to significant performance degradation for these methods. To address this issue, we propose an end-to-end semi-supervised Differentiated Auxiliary guided Network (DANet) to best use a few annotated videos. Except for the common supervised learning on a few annotated videos, the DANet also involves the knowledge of multiple pre-trained auxiliary networks to optimize the ViAR network in a self-supervised way on the unannotated data by removing the annotations. Considering the tight connection between video action recognition and classical static image-based visual tasks, the abundant knowledge from the pre-trained static image-based models can be used for training the ViAR model. Specifically, the DANet is a two-branch architecture, which includes a target branch of the ViAR network, and an auxiliary branch of multiple auxiliary networks (i.e., referring to diverse off-the-shelf models of relevant image tasks). Given a limited number of annotated videos, we train the target ViAR network end-to-end in a semi-supervised way, namely, with both the supervised cross-entropy loss on annotated videos, and the per-auxiliary weighted self-supervised contrastive losses on the same videos but without using annotations. Besides, we further explore different weighted guidance of the auxiliary networks to the ViAR network to better reflect different relationships between the image-based models and the ViAR model. Finally, we conduct extensive experiments on several popular action recognition benchmarks in comparison with existing state-of-the-art methods, and the experimental results demonstrate the superiority of DANet over most of the compared methods. In particular, the DANet obviously suppresses state-of-the-art ViAR methods even with very fewer annotated videos.

Multiparty Dual Learning.

The performance of machine learning algorithms heavily relies on the availability of a large amount of training data. However, in reality, data usually reside in distributed parties such as different institutions and may not be directly gathered and integrated due to various data policy constraints. As a result, some parties may suffer from insufficient data available for training machine learning models. In this article, we propose a multiparty dual learning (MPDL) framework to alleviate the problem of limited data with poor quality in an isolated party. Since the knowledge-sharing processes for multiple parties always emerge in dual forms, we show that dual learning is naturally suitable to handle the challenge of missing data, and explicitly exploits the probabilistic correlation and structural relationship between dual tasks to regularize the training process. We introduce a feature-oriented differential privacy with mathematical proof, in order to avoid possible privacy leakage of raw features in the dual inference process. The approach requires minimal modifications to the existing multiparty learning structure, and each party can build flexible and powerful models separately, whose accuracy is no less than nondistributed self-learning approaches. The MPDL framework achieves significant improvement compared with state-of-the-art multiparty learning methods, as we demonstrated through simulations on real-world datasets.

Semisupervised Graph Neural Networks for Graph Classification.

Graph classification aims to predict the label associated with a graph and is an important graph analytic task with widespread applications. Recently, graph neural networks (GNNs) have achieved state-of-the-art results on purely supervised graph classification by virtue of the powerful representation ability of neural networks. However, almost all of them ignore the fact that graph classification usually lacks reasonably sufficient labeled data in practical scenarios due to the inherent labeling difficulty caused by the high complexity of graph data. The existing semisupervised GNNs typically focus on the task of node classification and are incapable to deal with graph classification. To tackle the challenging but practically useful scenario, we propose a novel and general semisupervised GNN framework for graph classification, which takes full advantage of a slight amount of labeled graphs and abundant unlabeled graph data. In our framework, we train two GNNs as complementary views for collaboratively learning high-quality classifiers using both labeled and unlabeled graphs. To further exploit the view itself, we constantly select pseudo-labeled graph examples with high confidence from its own view for enlarging the labeled graph dataset and enhancing predictions on graphs. Furthermore, the proposed framework is investigated on two specific implementation regimes with a few labeled graphs and the extremely few labeled graphs, respectively. Extensive experimental results demonstrate the effectiveness of our proposed semisupervised GNN framework for graph classification on several benchmark datasets.

Deep-Growing Neural Network With Manifold Constraints for Hyperspectral Image Classification.

In the absence of sufficient labels, deep neural networks (DNNs) are prone to overfitting, resulting in poor performance and difficulty in training. Thus, many semisupervised methods aim to use unlabeled sample information to compensate for the lack of label quantity. However, as the available pseudolabels increase, the fixed structure of traditional models has difficulty in matching them, limiting their effectiveness. Therefore, a deep-growing neural network with manifold constraints (DGNN-MC) is proposed. It can deepen the corresponding network structure with the expansion of a high-quality pseudolabel pool and preserve the local structure between the original and high-dimensional data in semisupervised learning. First, the framework filters the output of the shallow network to obtain pseudolabeled samples with high confidence and adds them to the original training set to form a new pseudolabeled training set. Second, according to the size of the new training set, it increases the depth of the layers to obtain a deeper network and conducts the training. Finally, it obtains new pseudolabeled samples and deepens the layers again until the network growth is completed. The growing model proposed in this article can be applied to other multilayer networks, as their depth can be transformed. Taking HSI classification as an example, a natural semisupervised problem, the experimental results demonstrate the superiority and effectiveness of our method, which can mine more reliable information for better utilization and fully balance the growing amount of labeled data and network learning ability.

A Collaborative Multimodal Learning-Based Framework for COVID-19 Diagnosis.

The pandemic of coronavirus disease 2019 (COVID-19) has led to a global public health crisis, which caused millions of deaths and billions of infections, greatly increasing the pressure on medical resources. With the continuous emergence of viral mutations, developing automated tools for COVID-19 diagnosis is highly desired to assist the clinical diagnosis and reduce the tedious workload of image interpretation. However, medical images in a single site are usually of a limited amount or weakly labeled, while integrating data scattered around different institutions to build effective models is not allowed due to data policy restrictions. In this article, we propose a novel privacy-preserving cross-site framework for COVID-19 diagnosis with multimodal data, seeking to effectively leverage heterogeneous data from multiple parties while preserving patients' privacy. Specifically, a Siamese branched network is introduced as the backbone to capture inherent relationships across heterogeneous samples. The redesigned network is capable of handling semisupervised inputs in multimodalities and conducting task-specific training, in order to improve the model performance of various scenarios. The framework achieves significant improvement compared with state-of-the-art methods, as we demonstrate through extensive simulations on real-world datasets.