Singapore Dataset Research Articles

Graph deep learning recognition of port ship behavior patterns from a network approach

The detection and recognition of ship mobile behavior patterns based on automatic identification system (AIS) data are crucial for ship navigation and maritime supervision. However, present methods, such as rule-based constraints and density clustering extraction, still suffer from restrictions imposed by artificial settings of parameters and thresholds. The lack of neighborhood context information in unstructured and serialized ship trajectory vector data makes it challenging to incorporate advanced methods such as deep learning for pattern learning. This study builds multiple semi-supervised deep learning graph neural network (GNN) models based on the message passing paradigm for recognizing ship mobile behavior patterns by constructing an AIS trajectory network and computing multi-order node features through Delaunay triangulation. It can get rid of the dependence on parameters of traditional rule-based and unsupervised learning methods. It also reduces the dependence on auxiliary information in trajectory qualitative by introducing multi-order node features. We conducted the experiments on cargo ship data from New York and Singapore to test our method. The model's classification accuracy is more than 99% for the New York dataset and more than 95% for the Singapore dataset. This method can be extended to other types of ships and traffic trajectory vector data with the excellent application potential of GNN in pattern recognition of vector mobile data. The code and experimental data are available at: https://github.com/destiny1103/DT-GNN.

A long short-term memory-based hybrid model optimized using a genetic algorithm for particulate matter 2.5 prediction

Beijing, Shanghai, Singapore, and London are regions with high population density and industrial activities. In this sense, accurate prediction of the rate of particulate matter 2.5 (PM_2.5), one of the most important air pollutants, is very important for creating a healthy ecological system. This study proposes a hybrid method to predict PM_2.5 with high success. This method aims to optimize the Long Short-Term Memory (LSTM) deep learning model, which successfully solves time series problems with a Genetic Algorithm (GA). Here, the hyperparameters of the LSTM model, such as the number of layers, number of neurons, number of epochs, and kernel size, are optimized to achieve the highest prediction rate. Against this backdrop, the present study uses for the first time a hybrid GA-LSTM model to predict PM_2.5 concentrations in Beijing, Shanghai, Singapore, and London. The Beijing dataset consists of 13 attributes, the Shanghai dataset consists of 10 attributes, the London dataset consists of 5 attributes, and the Singapore dataset consists of 7 attributes. In this study, input variables such as PM_2.5, DEWP, TEMP, PRESS, WS, WD, PRE, Snowfall hours, Rain hours, and Season were used. To assess the effectiveness of this hybrid method, the developed GA-LSTM model was compared with k-Nearest Neighbors (kNN), Random Forest (RF), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), Convolutional Neural Networks (CNN), LSTM, and Gated Recurrent Units (GRU). Experimental results showed that GA-LSTM outperformed the compared models. In addition, one of the best results was obtained compared to other models in the literature.

Multi-fold Correlation Attention Network for Predicting Traffic Speeds with Heterogeneous Frequency

Short-term traffic prediction (e.g., less than 15 min) is challenging due to severe fluctuations of traffic data caused by dynamic traffic conditions and uncertainties (e.g., in data acquisition, driver behaviors, etc.). Substantial efforts have been undertaken to incorporate spatiotemporal correlations for improving traffic prediction accuracy. In this paper, we demonstrate that closely located road segments exhibit diverse spatial correlations when characterized using different measurements, and considering these multi-fold correlations can improve prediction performance. We propose new measurements to model multiple spatial correlations among traffic data. We develop a Multi-fold Correlation Attention Network (MCAN) that achieves accurate prediction by capturing multi-fold spatial correlation and multi-fold temporal correlations, and incorporating traffic data of heterogeneous sampling frequencies. The effectiveness of MCAN has been extensively evaluated on two real-world datasets in terms of overall performance, ablation study, sensitivity analysis, and case study, by comparing with several state-of-the-art methods. The results show that MCAN outperforms the best baseline with a reduction in mean absolute error (MAE) by 13% on Singapore dataset and 11% on Beijing dataset.

Automated Interpretation Of Systolic And Diastolic Function On The Echocardiogram

Background Echocardiography is the test of choice to assess cardiac function to diagnose and manage heart disease. However, manual interpretation of the echocardiogram is time-consuming and subject to human error. Therefore, we demonstrate a fully automated deep learning workflow to classify, segment, and annotate 2D videos and Doppler modalities. Methods The workflow was developed using a total of 20,828 annotated 2D and Doppler modalities images from 1,145 studies (N) from a heart failure research platform in Singapore. We validated the workflow against human measurements in a dataset from Canada (N=1,029), a 'real-world' dataset from Taiwan (N=31,241), the US-based EchoNet-Dynamic dataset (N=10,030); and in an independent prospective Singapore dataset (N=29 to 66) with repeated human expert measurements. Results In the test set, the automated workflow classified 2D videos and Doppler modalities with accuracies ranging from 0.91 to 0.99. The left ventricle and left atrium segmentation were accurate, with a mean Dice similarity coefficient >93% for all. In the external datasets, automated measurements showed good agreement with locally measured values, with a mean absolute error range of 13-25ml for left ventricular volumes, 6-10% for left ventricular ejection fraction, and 1.8-2.0 for E/e'; and reliably classified systolic dysfunction (LVEF<40%, area under the curve [AUC] range 0.90-0.92, Figure A) and diastolic dysfunction (E/e' ≥13, AUC range 0.91-0.91, Figure B). Independent prospective evaluation confirmed similar or less variance of automated compared to human expert measurements. Conclusions : Deep learning algorithms can automatically annotate 2D videos and Doppler modalities with similar accuracy compared with human measurements. Echocardiography is the test of choice to assess cardiac function to diagnose and manage heart disease. However, manual interpretation of the echocardiogram is time-consuming and subject to human error. Therefore, we demonstrate a fully automated deep learning workflow to classify, segment, and annotate 2D videos and Doppler modalities. The workflow was developed using a total of 20,828 annotated 2D and Doppler modalities images from 1,145 studies (N) from a heart failure research platform in Singapore. We validated the workflow against human measurements in a dataset from Canada (N=1,029), a 'real-world' dataset from Taiwan (N=31,241), the US-based EchoNet-Dynamic dataset (N=10,030); and in an independent prospective Singapore dataset (N=29 to 66) with repeated human expert measurements. In the test set, the automated workflow classified 2D videos and Doppler modalities with accuracies ranging from 0.91 to 0.99. The left ventricle and left atrium segmentation were accurate, with a mean Dice similarity coefficient >93% for all. In the external datasets, automated measurements showed good agreement with locally measured values, with a mean absolute error range of 13-25ml for left ventricular volumes, 6-10% for left ventricular ejection fraction, and 1.8-2.0 for E/e'; and reliably classified systolic dysfunction (LVEF<40%, area under the curve [AUC] range 0.90-0.92, Figure A) and diastolic dysfunction (E/e' ≥13, AUC range 0.91-0.91, Figure B). Independent prospective evaluation confirmed similar or less variance of automated compared to human expert measurements. : Deep learning algorithms can automatically annotate 2D videos and Doppler modalities with similar accuracy compared with human measurements.

Image-Based Geo-Localization Using Satellite Imagery

The problem of localization on a geo-referenced satellite map given a query ground view image is useful yet remains challenging due to the drastic change in viewpoint. To this end, in this paper we work on the extension of our earlier work on the Cross-View Matching Network (CVM-Net) (Hu et al. in IEEE conference on computer vision and pattern recognition (CVPR), 2018) for the ground-to-aerial image matching task since the traditional image descriptors fail due to the drastic viewpoint change. In particular, we show more extensive experimental results and analyses of the network architecture on our CVM-Net. Furthermore, we propose a Markov localization framework that enforces the temporal consistency between image frames to enhance the geo-localization results in the case where a video stream of ground view images is available. Experimental results show that our proposed Markov localization framework can continuously localize the vehicle within a small error on our Singapore dataset.

Voronoi-Visibility Roadmap-based Path Planning Algorithm for Unmanned Surface Vehicles

In this paper, a novel Voronoi-Visibility (VV) path planning algorithm, which integrates the merits of a Voronoi diagram and a Visibility graph, is proposed for solving the Unmanned Surface Vehicle (USV) path planning problem. The VM (Voronoi shortest path refined by Minimising the number of waypoints) algorithm was applied for performance comparison. The VV and VM algorithms were compared in ten Singapore Strait missions and five Croatian missions. To test the computational time, a high-resolution, large spatial dataset was used. It was demonstrated that the proposed algorithm not only improved the quality of the Voronoi shortest path but also maintained the computational efficiency of the Voronoi diagram in dealing with different geographical scenarios, while also keeping the USV at a configurable clearance distance c from coastlines. Quantitative results were generated by comparing the Voronoi, VM and VV algorithms in 2,000 randomly generated missions using the Singapore dataset.

Identification of ARL4C as a Peritoneal Dissemination-Associated Gene and Its Clinical Significance in Gastric Cancer.

In gastric cancer (GC), peritoneal dissemination (PD) occurs frequently and is incurable. In this study, we aimed to identify PD-associated genes in GC. We identified a PD-associated gene using three GC datasets: highly disseminated peritoneal GC cell lines, the Singapore dataset and The Cancer Genome Atlas (TCGA) dataset. We assessed the clinicopathological significance of the gene expression using reverse transcription-quantitative polymerase chain reaction (RT-qPCR) and performed immunohistochemical analysis for the gene in our patient cohort. We also performed survival analyses of the gene in our patient cohort, the Singapore dataset and the GSE62254 datasets. Moreover, gene set enrichment analysis (GSEA) was performed using the Singapore and TCGA datasets. Finally, in vitro experiments such as invasion/migration assays, immunofluorescence staining of actin filaments, epidermal growth factor (EGF) treatment analysis, and gene expression analysis were conducted using three gene-knockdown GC cell lines (AGS, 58As9, MKN45). ADP-ribosylation factor-like 4c (ARL4C) was identified as a PD-associated gene, and immunohistochemical analysis showed that ARL4C was overexpressed in GC cells. High ARL4C expression was associated with the depth of invasion (p<0.01) and PD (p<0.05) and was a poor prognostic factor (p<0.05) in our patient cohort, the Singapore dataset and the GSE62254 dataset. ARL4C expression positively correlated with the epithelial-mesenchymal transition (EMT) gene set in GSEA. Moreover, ARL4C knockdown reduced invasion/migration capacity, SLUG expression, and the formation of lamellipodia or filopodia in AGS and 58As9 cells. Finally, EGF treatment increased ARL4C expression in MKN45 cells. ARL4C was associated with PD and was a poor prognostic factor in GC, possibly through promoting invasive capacity by activation of both EMT and motility.

Validation of the 18-gene classifier as a prognostic biomarker of distant metastasis in breast cancer.

We validated an 18-gene classifier (GC) initially developed to predict local/regional recurrence after mastectomy in estimating distant metastasis risk. The 18-gene scoring algorithm defines scores as: <21, low risk; ≥21, high risk. Six hundred eighty-three patients with primary operable breast cancer and fresh frozen tumor tissues available were included. The primary outcome was the 5-year probability of freedom from distant metastasis (DMFP). Two external datasets were used to test the predictive accuracy of 18-GC. The 5-year rates of DMFP for patients classified as low-risk (n = 146, 21.7%) and high-risk (n = 537, 78.6%) were 96.2% (95% CI, 91.1%–98.8%) and 80.9% (74.6%–81.9%), respectively (median follow-up interval, 71.8 months). The 5-year rates of DMFP of the low-risk group in stage I (n = 62, 35.6%), stage II (n = 66, 20.1%), and stage III (n = 18, 10.3%) were 100%, 94.2% (78.5%–98.5%), and 90.9% (50.8%–98.7%), respectively. Multivariate analysis revealed that 18-GC is an independent prognostic factor of distant metastasis (adjusted hazard ratio, 5.1; 95% CI, 1.8–14.1; p = 0.0017) for scores of ≥21. External validation showed that the 5-year rate of DMFP in the low- and high-risk patients was 94.1% (82.9%–100%) and 80.3% (70.7%–89.9%, p = 0.06) in a Singapore dataset, and 89.5% (81.9%–94.1%) and 73.6% (67.2%–79.0%, p = 0.0039) in the GEO-GSE20685 dataset, respectively. In conclusion, 18-GC is a viable prognostic biomarker for breast cancer to estimate distant metastasis risk.

Two-Level Feature Representation for Aerial Scene Classification

Effective scene representation is a fundamental part of high-resolution scene classification systems. In this letter, we present a holistic scene representation method, i.e., the two-level feature representation (TLFR) model. The TLFR is composed of low-level and high-level features. Low-level features are obtained by computing the residual error between a local descriptor and its corresponding visual word, and the high-level features are obtained using a proposed selection-constrained sparse coding method. In addition, low-level features in a cluster are integrated by summation pooling, whereas high-level features are fused by maximization pooling. The holistic scene representation is finally generated by incorporating these two levels of features into the bag-of-visual-words framework. Experimental results show that the TLFR model is robust to translation and rotation variations and demonstrates promising performance with the Land Use and Land Cover Database data set and a newly released Singapore data set.

Abstract B22: Integrative analysis of DNA methylation and gene expression data reveals complex regulation of gastric cancer

Abstract Gastric cancer is a heterogeneous disease where diverse genetic and epigenetic alternations can accumulate in different molecular and histological subtypes. We applied our recently developed causality test between genome-wide DNA methylation and gene expression profiles to three independent cohorts of gastric tumors (97 in Hong Kong University (HKU), 159 in University of Singapore (Singapore) and 365 samples in TCGA stomach adenocarcinoma (TCGA) ). We focused on methylation variations within CpG islands in promoter regions, where global hypermethylation was observed, and identified 37, 62, and 537 key regulators in HKU, Singapore, and TCGA dataset respectively. There were 5 common key regulators (ADHFE1, CDO1, COX7A1, FSTL1, and TCF21) whose methylation variations had high impact on mRNA level changes of large number of downstream genes in all three dataset where different cohorts were profiled using different platforms. When we compared two dataset, there were 5 common key regulators in HKU and Singapore dataset (Fisher's exact test (FET) p-value = 2.9×e-07), 27 common key regulators in HKU and TCGA dataset (FET p-value = 6.8×e-35), and 30 common key regulators in Singapore and TCGA dataset (FET p-value = 1.0×e-29). By combining these, 52 genes were identified as key regulators within at least two dataset. Several of the key regulators were known for the association between their epigenetic disruption and the disease (for example, BNIP3, CDO1, TCF21, ZSCAN18, and so on) while other genes have not implicated in the gastric cancer previously. More interestingly, the downstream genes of these key regulators were significantly overlapped and the directions of correlation with methylation levels were almost same within the three dataset. Further clustering key regulators based on their downstream genes overlaps revealed that there were two distinct groups of downstream genes commonly regulated by these key regulators and the expression of these two groups were anti-correlated. One group was enriched for cell cycle related genes and the other group was enriched for genes involved in immune responses. This result indicates that cell cycle and immune response functions were inversely regulated by methylation variations of the same set of genes. It is worth to note that methylation patterns of some key regulators were subtype dependent and the subtype specific methylation patterns were only observed in tumor samples, but not in adjacent normal tissues. Based on integrative analysis of genome-wide DNA methylation and gene expression profiles within three independent gastric cancer dataset, we identified a set of key regulators whose methylation changes might play a ‘causal' role in the transcriptional regulation associated with the gastric cancer. Further experiments are needed to validate and dissect these putative candidate genes' roles in tumorigenesis and progression of this complex and heterogeneous disease. Citation Format: Seungyeul Yoo, Suet Yi Leung, Jun Zhu. Integrative analysis of DNA methylation and gene expression data reveals complex regulation of gastric cancer. [abstract]. In: Proceedings of the AACR Special Conference on Chromatin and Epigenetics in Cancer; Sep 24-27, 2015; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2016;76(2 Suppl):Abstract nr B22.

Radar measured rain attenuation with proposed Z–R relationship at a tropical location

Attenuation measurements from two Ku- and a Ka-band satellite beacon signals in a tropical site, Singapore are presented. A new Z–R relationship (Z=61.75·R1.61) is derived using rainfall rate data from rain gauge. Rainfall rates predicted from Radar data using the proposed relationship and the Singapore data set are used to compute the earth–satellite path rain attenuation. The measured attenuations from satellite beacon receivers match well with that computed values using the Radar data.

Long-Term Effect of Curing Temperature on the Strength Behavior of Cement-Stabilized Clay

The influence of curing temperature on the strength development of concrete, mortar, or cement-stabilized granular soils has been widely reported in the literature. However, there are fewer studies focusing on cement-stabilized clay. The key difference is the presence of pozzolanic reactions in cement-stabilized clay, which give rise to more complex strength-development behaviors than those of concrete, mortar, or cement-stabilized granular soils. This paper aims to clarify this difference using extensive data collected as part of a land reclamation project in Singapore. Five different mixes of cement-stabilized Singapore marine clay are cured under different temperatures. Their strength development with time is studied by testing the specimens under unconfined compression at various curing ages (up to 1 year). The experimental results show that a higher curing temperature gives rise to not only higher short-term strengths but also higher long-term/ultimate strengths. Based on the experimental results, a model is proposed that combines the maturity theory commonly used for concrete or mortar with a proposed temperature-enhanced strength factor, which takes into account the effect of curing temperature on pozzolanic reactions occurring in cement-stabilized clay. This proposed model is validated by another independent Singapore data set and data sets collected from the literature.

TROPICAL RAIN CLASSIFICATION AND ESTIMATION OF RAIN FROM Z-R (REFLECTIVITY-RAIN RATE) RELATIONSHIPS

A Z-R relation is derived using a data set which consists of nine rain events selected from Singapore's drop size distribution. Rain events are separated into convective and stratiform types of rain using two methods: the Gamache-Houze method, a simple threshold technique, and the Atlas-Ulbrich method. In the Atlas-Ulbrich method, the variability of the rain integral parameters R; Z, Nw, D0 and gamma model parameter are used for the classiflcation of rain into convective, stratiform and transition. Z-R relations are derived for each type of rain after classiflcation. The changes in the coe-cients of the Z-R relations for difierent rain events are plotted and analyzed. The Z-R relations of the difierent methods using the Singapore data are compared and analyzed. It is concluded that the coe-cient A of the Z-R relation is higher for the convective stage followed by the stratiform and transition stages. The coe-cient b values are higher for the transition stage followed by the stratiform and convective stages. Re∞ectivities are extracted from RADAR data above NTU site for rain events and compared with the re∞ectivities derived from the distrometer data. Rain rates retrieved from RADAR data using the proposed relations from Singapore's data set are compared with the distrometer rain rates. The RADAR extracted rain rates are found to be constantly lower than the distrometer derived rain rates but matches well.