Non-matching Pairs Research Articles

Privately evaluating sensitive population record linkage without ground truth data

AbstractData containing personal information allow detailed studies in the health and social sciences, such as population-related analysis. However, such studies often require the linking of two or more databases because information about a person can be scattered across multiple data sources. To address this issue of data being scattered, researchers have been working on linking records across multiple data sources to identify records that refer to the same person, or the same group of individuals (known as group linkage) using quasi-identifiers such as names and addresses which can be missing, out of date or contain errors or variations, making record linkage a very challenging task. Record linkage applications often also lack ground truth data in the form of matching and non-matching record pairs, which challenges the assessment of the quality of linkage algorithms. Furthermore, when linkage is conducted on sensitive data, for example personal health records, due to privacy concerns ground truth can generally not be obtained using methods such as crowd sourcing. This study therefore aims to develop methods to assess the linkage quality of sensitive data by using publicly available data sets, such as census or voter data, in a privacy-preserving manner, with a focus on the group linkage problem. Assuming that distinct groups, such as siblings in a family, are identifiable in both the sensitive and public data sets, we develop a novel method to estimate linkage quality using public data by encoding information that is commonly available in both sensitive and public data sets into a common representation using Bloom filters. Comparing these Bloom filters then allows the estimation of linkage quality. An evaluation using a real sensitive birth data set and a public census data set from Scotland shows the effectiveness of our proposed method for quality estimation, which achieves a median correlation of 98% with linkage quality calculated based on ground truth data.

Matched vs Nonmatched Placebos in a Randomized Trial of COVID-19 Treatments

Matched placebo interventions are complex and resource intensive. Recent evidence suggests matched placebos may not always be necessary. Previous studies have predominantly evaluated potential bias of nonmatched placebos (ie, differing on dose, frequency of administration, or formulation) in pain and mental health, but to date no systematic examination has been conducted in infectious disease. To test for differences between nonmatched and matched placebo arms with respect to clinical outcome measures across multiple therapeutics for COVID-19. In a comparative effectiveness research study, a post hoc analysis was conducted of data on individual patients enrolled in a large, multiarm, platform randomized clinical trial in symptomatic adult outpatients with COVID-19 between January 15, 2021, to September 28, 2023, in which the outcomes of both matched and nonmatched placebo groups were reported. Bayesian and frequentist covariate-adjusted techniques were compared with 7 intervention-placebo pairs. Seven matched and nonmatched placebo pairs (for a total of 7 comparisons) were evaluated throughout the primary platform trial. Comparisons were made between treatment and its associated matched (concurrent) placebo, as well as with nonmatched placebo (alone and in combination) assessed at a similar time point. Outcomes assessed included hospitalizations, EuroQol 5-Dimension 5-level scores, and PROMIS Global-10 scores. A total of 7 intervention-control pairs (N = 2684) were assessed, including 1620 (60.4%) women, with mean (SD) age, 47 (15.2) years; the most common comorbidities were obesity (41.9%) and hypertension (37.9%). In a meta-analysis with decoupled SEs, accounting for overlapping placebo patients, the overall odds ratio (OR) of nonmatched compared with matched placebo was 1.01 (95% credible interval, 0.77-1.32), with posterior probability of equivalence, defined as 0.8 ≤ OR ≤ 1.2 (a deviation from perfect equivalence ie, OR = 1, by no more than 0.2) of 85.4%, implying no significant difference. Unadjusted analysis of the event rate difference between all nonmatched and matched placebo groups did not identify any notable differences across all 7 treatment-placebo combinations assessed. Similar analysis that was conducted for patient-reported quality of life outcomes did not yield statistically significant differences. In this post hoc study of a randomized clinical platform trial, pooling matched and nonmatched placebo patient data did not lead to inconsistencies in treatment effect estimation for any of the investigational drugs. These findings may have significant implications for future platform trials, as the use of nonmatched placebo may improve statistical power, or reduce barriers to placebo implementation.

Cancelable templates for secure face verification based on deep learning and random projections

Recently, biometric recognition has become a significant field of research. The concept of cancelable biometrics (CB) has been introduced to address security concerns related to the handling of sensitive data. In this paper, we address unconstrained face verification by proposing a deep cancelable framework called BiometricNet+ that employs random projections (RP) to conceal face images and compressive sensing (CS) to reconstruct measurements in the original domain. Our lightweight design enforces the properties of unlinkability, revocability, and non-invertibility of the templates while preserving face recognition accuracy. We compare facial features by learning a regularized metric: at training time, we jointly learn facial features and the metric such that matching and non-matching pairs are mapped onto latent target distributions; then, for biometric verification, features are randomly projected via random matrices changed at every enrollment and query and reconstructed before the latent space mapping is computed. We assess the face recognition accuracy of our framework on challenging datasets such as LFW, CALFW, CPLFW, AgeDB, YTF, CFP, and RFW, showing notable improvements over state-of-the-art techniques while meeting the criteria for secure cancelable template design. Since our method requires no fine-tuning of the learned features, it can be applied to pre-trained networks to increase sensitive data protection.

A lattice structure for ancestral configurations arising from the relationship between gene trees and species trees

To a given gene tree topology G and species tree topology S with leaves labeled bijectively from a fixed set X, one can associate a set of ancestral configurations, each of which encodes a set of gene lineages that can be found at a given node of the species tree. We introduce a lattice structure on ancestral configurations, studying the directed graphs that provide graphical representations of lattices of ancestral configurations. For a matching gene tree topology and species tree topology G=S, we present a method for defining the digraph of ancestral configurations from the tree topology by using iterated cartesian products of graphs. We show that a specific set of paths on the digraph of ancestral configurations is in bijection with the set of labeled histories — a well-known phylogenetic object that enumerates possible temporal orderings of the coalescences of a tree. For each of a series of tree families, we obtain closed-form expressions for the number of labeled histories by using this bijection to count paths on associated digraphs. Finally, we prove that our lattice construction extends to nonmatching tree pairs, and we use it to characterize pairs (G,S) having the maximal number of ancestral configurations for a fixed G. We discuss how the construction provides new methods for performing enumerations of combinatorial aspects of gene and species trees.

The low prevalence effect in fingerprint comparison amongst forensic science trainees and novices.

The low prevalence effect is a phenomenon whereby target prevalence affects performance in visual search (e.g., baggage screening) and comparison (e.g., fingerprint examination) tasks, such that people more often fail to detect infrequent target stimuli. For example, when exposed to higher base-rates of ‘matching’ (i.e., from the same person) than ‘non-matching’ (i.e., from different people) fingerprint pairs, people more often misjudge ‘non-matching’ pairs as ‘matches’–an error that can falsely implicate an innocent person for a crime they did not commit. In this paper, we investigated whether forensic science training may mitigate the low prevalence effect in fingerprint comparison. Forensic science trainees (n = 111) and untrained novices (n = 114) judged 100 fingerprint pairs as ‘matches’ or ‘non-matches’ where the matching pair occurrence was either high (90%) or equal (50%). Some participants were also asked to use a novel feature-comparison strategy as a potential attenuation technique for the low prevalence effect. Regardless of strategy, both trainees and novices were susceptible to the effect, such that they more often misjudged non-matching pairs as matches when non-matches were rare. These results support the robust nature of the low prevalence effect in visual comparison and have important applied implications for forensic decision-making in the criminal justice system.

DADER

Entity resolution (ER) is a core data integration problem that identifies pairs of data instances referring to the same real-world entities, and the state-of-the-art results of ER are achieved by deep learning (DL) based approaches. However, DL-based approaches typically require a large amount of labeled training data (i.e. , matching and non-matching pairs), which incurs substantial manual labeling efforts. In this paper, we introduce DADER , a hands-off deep ER system through domain adaptation. DADER utilizes multiple well-labeled source ER datasets to train a DL-based ER model for a new target ER dataset that does not have any labels or with only a few labels. To address the key challenge of domain shift, DADER judiciously selects labeled entity pairs from the source and then aligns distributions of the source and the target by using six popular domain adaptation strategies. DADER can also harness the users to gather a few labels for further improvement. We have built DADER as an open-sourced Python Library with intuitive APIs and demonstrated its utility on supporting hands-off ER in real-world scenarios.

Virtual image standard (VIS) for performance evaluation of the congruent matching cells (CMC) algorithms in firearm evidence identifications

The congruent matching cells (CMC) method was invented at the National Institute of Standards and Technology (NIST) in 2012 for automatic and objective firearm evidence identifications and estimation of the weight of evidence in firearm evidence identifications. Since 2013, five CMC algorithms have been developed at NIST. In this paper, the virtual image standard (VIS) is proposed through trimming and stitching KNM images for quantitative performance evaluations of different CMC algorithms. The evaluation criteria include the correlation accuracy (both the CMC numbers and distribution pattern), correlation efficiency, false positive (FP) error rate, and the maximum separation of known matching (KM) and known non-matching (KNM) image pairs. The VIS composes correlation cells from different KNM images, which can provide a ground truth for verifying the CMC numbers, distribution patterns, and FP errors. By identifying three groups of VIS, the Convergence CMC algorithm showed superior performances for the future casework in firearm evidence identifications. Lastly, the success of this study suggests that the VIS could also be used to optimize the correlation parameters, to develop and test new CMC algorithms, and evaluate the performance before it is put into use for firearm examiner's casework.

Approach to breech face impression comparison based on the robust estimation of a correspondence function

Forensic firearm analysis concerns an attempt to determine if ammunition is associated with a specific firearm based on tool-marks produced by it. A feature-based method using the Scale Invariant Feature Transform (SIFT) and RANdom SAmple Consensus (RANSAC) integration algorithm had been suggested to allow the automated comparison of breech face impressions. In this paper, an estimation method is proposed to establish a correspondence function among the features of comparison impression pairs, aiming to further improve the robustness and repeatability of automated feature matching. During the application of the iterative establishment algorithm, the Support Vector Regression (SVR) method is repeated to estimate the correspondence function based on current feature correspondences, and a robust weighting method excludes egregious outliers among putative correspondences by updating additional weightings. Moreover, the consistency detection method is adopted to overcome the over-fitting problem in SVR. Validation tests of the proposed method are conducted on three sets of cartridge case’s breech face impressions; namely the Fadul set consisting of 40 cartridge cases ejected from 10 Ruger P95PR15 pistols, the Weller sets containing 95 cartridge cases obtained from 11 Ruger P95DC firearms and the Lightstone set containing 30 cartridge cases from 10 SW40VE S&W Sigma pistol slides. Test results show that most known matching (KM) pairs possess no less than 20 matching feature points while the non-matching (KNM) pairs all maintain 3–8 correspondences. It also indicates that the feature-based method has apparent advantages in dealing with granular impressions with local peaks and valleys features, and poor performance on the striation marks. The clear distinction between KM and KNM impression pairs demonstrates the feasibility of the proposed method in ballistic feature comparison. Compared to the random hypothesize-and-verify modeling of RANSAC, this method can retain more reliable matching feature points of the impression pair to ensure the repeatability of feature correspondence selection.

Similarity evaluation of 3D surface topography measurements

With the recent advances in three-dimensional (3D) optical scanning technologies, 3D surface topography measurement plays an increasingly important role in many fields, such as product quality inspection in additive manufacturing (AM), gauge capability analysis, and firearm identification in forensic science. In this paper, we establish a thorough and flexible new framework of surface topography data comparison that generates a scaled similarity score for a pair of measurements, and distinguishes matched and non-matched pairs based on the score. If two measurements are portraying the same surface, they are defined as a matched pair. If not, they are defined as a non-matched pair. This similarity evaluation framework can be very useful in comparing optical scanning systems, quantifying different sources of variation in a process, and monitoring the stability and uniformity of a process, thus has a great potential to improve the quality assurance cycle of AM processes in the long run. We illustrate the framework and statistically evaluate the binary classification performance on data measured from additive manufactured parts on a large scale. We also examine how different systems, repeated measurements, and operators affect the similarity score and classification performance. We compare our work with a baseline method using the surface roughness average parameter Ra. The results show that the methodology can distinguish matched and non-matched pairs with high accuracy, and it outperforms the baseline method greatly. The framework serves as a benchmark method and can be generalized to be used in other fields where surface topography plays a critical role.

Enumeration of coalescent histories for caterpillar species trees and p-pseudocaterpillar gene trees

For a fixed set X containing n taxon labels, an ordered pair consisting of a gene tree topology G and a species tree topology S bijectively labeled with the labels of X possesses a set of coalescent histories—mappings from the set of internal nodes of G to the set of edges of S describing possible lists of edges in S on which the coalescences in G take place. Enumerations of coalescent histories for gene trees and species trees have produced suggestive results regarding the pairs (G,S) that, for a fixed n, have the largest number of coalescent histories. We define a class of 2-cherry binary tree topologies that we term p-pseudocaterpillars, examining coalescent histories for non-matching pairs (G,S) in the case in which S has a caterpillar shape and G has a p-pseudocaterpillar shape. Using a construction that associates coalescent histories for (G,S) with a class of “roadblocked” monotonic paths, we identify the p-pseudocaterpillar labeled gene tree topology that, for a fixed caterpillar labeled species tree topology, gives rise to the largest number of coalescent histories. The shape that maximizes the number of coalescent histories places the “second” cherry of the p-pseudocaterpillar equidistantly from the root of the “first” cherry and from the tree root. A symmetry in the numbers of coalescent histories for p-pseudocaterpillar gene trees and caterpillar species trees is seen to exist around the maximizing value of the parameter p. The results provide insight into the factors that influence the number of coalescent histories possible for a given gene tree and species tree.

An improved subpixel-level registration method for image-based fault diagnosis of train bodies using SURF features

Using computer-vision-based fault-diagnosis technologies, anomalies in train bodies can be automatically identified as part of daily maintenance. Nevertheless, since the train speed changes constantly and nonlinearly during the process of image acquisition, the obtained images can be severely distorted. As such, it is necessary to first register the collected images to correct the distortion, so as to accurately extract the key parts of the train body for fault detection. This paper proposes a new accurate registration method for train images based on speeded-up robust features, which improves on the traditional segment registration (SR) method by adding procedures for feature selection, piecewise linearization, and interpolated registration. The novelty of the proposed method is that up to subpixel-level registration accuracy can be achieved. Meanwhile, the robustness of the algorithm is ensured by its strong ability to screen out non-matching pairs when selecting features for registration. To demonstrate the performance of the proposed scheme, experimental comparison and verification are conducted, which show that, compared with the traditional SR method, the registration accuracy is remarkably improved in terms of mean square error, normalization cross-correlation, and structural similarity index.

Demonstration of panda

Entity matching (EM) refers to the problem of identifying tuple pairs in one or more relations that refer to the same real world entities. Supervised machine learning (ML) approaches, and deep learning based approaches in particular, typically achieve state-of-the-art matching results. However, these approaches require many labeled examples, in the form of matching and non-matching pairs, which are expensive and time-consuming to label. In this paper, we introduce Panda, a weakly supervised system specifically designed for EM. Panda uses the same labeling function abstraction as Snorkel, where labeling functions (LF) are user-provided programs that can generate large amounts of (somewhat noisy) labels quickly and cheaply, which can then be combined via a labeling model to generate accurate final predictions. To support users developing LFs for EM, Panda provides an integrated development environment (IDE) that lives in a modern browser architecture. Panda's IDE facilitates the development, debugging, and life-cycle management of LFs in the context of EM tasks, similar to how IDEs such as Visual Studio or Eclipse excel in general-purpose programming. Panda's IDE includes many novel features purpose-built for EM, such as smart data sampling, a builtin library of EM utility functions, automatically generated LFs, visual debugging of LFs, and finally, an EM-specific labeling model. We show in this demo that Panda IDE can greatly accelerate the development of high-quality EM solutions using weak supervision.

The Presence of Recipient CMV Immunodominant Human Leukocyte Antigen (HLA) Alleles Affect Outcomes after Haploidentical (HI) Hematopoietic Stem Cell Transplantation (HSCT)

CMV reactivation is associated with increased treatment-related mortality (TRM), decreased overall survival (OS), and in some analyses, protection from relapse after HSCT. We previously showed that CMV reactivation was significantly higher in HI versus matched related HSCT recipients receiving an identical preparatory regimen, implicating HLA mismatch as a contributor to reactivation. It is known that specific CMV immunodominant HLA alleles (CMV ID alleles) are associated with more efficient antigen presentation and protection from CMV. We hypothesized that a match at a CMV ID allele in the shared haplotype of the donor and recipient pairs may overcome the disadvantage proffered by HI HSCT with respect to CMV. As a control, we tested whether the presence of a CMV ID allele in either the donor or recipient, as opposed to a match at the CMV ID allele, was the more important impactor of outcomes. This information would potentially impact donor selection and identify HI recipients with greater risk for CMV-related complications.MethodsThis is a retrospective, analysis of 138 HI HSCT recipients at risk for CMV reactivation as defined by the presence of positive CMV serologic status in the recipient, donor, or both. All recipients were treated on the Jefferson 2 step approach in which after conditioning, a fixed T cell dose (2 x 108/kg) is administered (step 1), followed 2-3 days later by cyclophosphamide (CY) for bidirectional tolerization. Two days after CY, patients receive a CD34-selected donor product (step 2). To be included in the analysis, patients had to successfully engraft and be alive for at least 1 month post HSCT. Outcomes related to CMV, 1) reactivation incidence, 2) peak copy number of initial CMV reactivation occurring by d+100, 3) TRM, 4) relapse related mortality (RRM), and 5) OS, were tested in 3 contexts: the presence vs the absence of a CMV ID allele in the recipient or the donor, or the presence or absence of a donor/recipient (R/D) match at the same CMV ID allele. CMV ID alleles were defined as HLA A01:01, A02:01, A24:02, B07:02, B08:01, B35:01 and C07:02. Statistics used: incidence of CMV reactivation=Chi-Square; comparison of peak CMV copy numbers=T-test; assessment of RRM and NRM=cumulative incidence (CI); OS=Kaplan-Meier method.Results:A CMV ID allele was present in 68.1% recipients and 66.7% donors. 41.7% of D/R pairs were matched at the same CMV ID allele. The incidence of CMV reactivation was not associated with a D/R CMV ID allele match or recipient or donor CMV ID allele positivity. Peak CMV levels were lower in the presence of any recipient or donor CMV ID allele, but only approached significance in D/R pairs matched at the same CMV ID allele (p=0.065) with a mean of 8.7 x 10e3 vs 4.1 x10e4 copies/ml by PCR in matched vs non-matched pairs respectively. At 3 years, in recipients who matched their donor at 1 or more CMV ID alleles vs those who had no match, CI of NRM was 28% vs 26% (p=0.48) and 16% vs 32% (p=0.07) for RRM. In recipients with vs without a CMV ID allele CI of NRM was 23% vs 36% (p=0.059) and 20% vs 36% for RRM (p= 0.02) (Figure). In recipients with donors who possessed a CMV ID allele vs not, CI of NRM was 27% vs 28% (p=0.80) and 21% vs 31% for RRM (p= 0.12). OS in recipients who were matched at the same CMV ID allele with their donors or who possessed a CMV ID allele was 51% vs 37% (p=0.062) and 52% vs 27% (p=0.001) respectively as compared to recipients negative for these alleles.Conclusion:This analysis suggests that matching HI D/R pairs at a CMV ID epitope may reduce CMV peak copy number but has no significant effect on post HSCT reactivation. Recipient positivity for a CMV ID allele was strongly associated with improvement in NRM, RRM, and OS. Whether the benefit is specific to improved CMV control or to a more efficient antigen presentation in general by these alleles is unknown, but should be further investigated. This data requires confirmation with multivariate analysis. [Display omitted] DisclosuresWagner:Novartis: Research Funding.

Efficient and effective ER with progressive blocking

Blocking is a mechanism to improve the efficiency of Entity Resolution (ER) which aims to quickly prune out all non-matching record pairs. However, depending on the distributions of entity cluster sizes, existing techniques can be either (a) too aggressive, such that they help scale but can adversely affect the ER effectiveness, or (b) too permissive, potentially harming ER efficiency. In this paper, we propose a new methodology of progressive blocking (pBlocking) to enable both efficient and effective ER, which works seamlessly across different entity cluster size distributions. pBlocking is based on the insight that the effectiveness-efficiency trade-off is revealed only when the output of ER starts to be available. Hence, pBlocking leverages partial ER output in a feedback loop to refine the blocking result in a data-driven fashion. Specifically, we bootstrap pBlocking with traditional blocking methods and progressively improve the building and scoring of blocks until we get the desired trade-off, leveraging a limited amount of ER results as a guidance at every round. We formally prove that pBlocking converges efficiently ($O(n log^2 n)$ time complexity, where n is the total number of records). Our experiments show that incorporating partial ER output in a feedback loop can improve the efficiency and effectiveness of blocking by 5x and 60% respectively, improving the overall F-score of the entire ER process up to 60%.

Ethnicity-based name partitioning for author name disambiguation using supervised machine learning.

In several author name disambiguation studies, some ethnic name groups such as East Asian names are reported to be more difficult to disambiguate than others. This implies that disambiguation approaches might be improved if ethnic name groups are distinguished before disambiguation. We explore the potential of ethnic name partitioning by comparing performance of four machine learning algorithms trained and tested on the entire data or specifically on individual name groups. Results show that ethnicity‐based name partitioning can substantially improve disambiguation performance because the individual models are better suited for their respective name group. The improvements occur across all ethnic name groups with different magnitudes. Performance gains in predicting matched name pairs outweigh losses in predicting nonmatched pairs. Feature (e.g., coauthor name) similarities of name pairs vary across ethnic name groups. Such differences may enable the development of ethnicity‐specific feature weights to improve prediction for specific ethic name categories. These findings are observed for three labeled data with a natural distribution of problem sizes as well as one in which all ethnic name groups are controlled for the same sizes of ambiguous names. This study is expected to motive scholars to group author names based on ethnicity prior to disambiguation.

NCNet: Neighbourhood Consensus Networks for Estimating Image Correspondences

We address the problem of finding reliable dense correspondences between a pair of images. This is a challenging task due to strong appearance differences between the corresponding scene elements and ambiguities generated by repetitive patterns. The contributions of this work are threefold. First, inspired by the classic idea of disambiguating feature matches using semi-local constraints, we develop an end-to-end trainable convolutional neural network architecture that identifies sets of spatially consistent matches by analyzing neighbourhood consensus patterns in the 4D space of all possible correspondences between a pair of images without the need for a global geometric model. Second, we demonstrate that the model can be trained effectively from weak supervision in the form of matching and non-matching image pairs without the need for costly manual annotation of point to point correspondences. Third, we show the proposed neighbourhood consensus network can be applied to a range of matching tasks including both category- and instance-level matching, obtaining the state-of-the-art results on the PF, TSS, InLoc, and HPatches benchmarks.

Examining the Dorsolateral and Ventromedial Prefrontal Cortex Involvement in the Self-Attention Network: A Randomized, Sham-Controlled, Parallel Group, Double-Blind, and Multichannel HD-tDCS Study.

BackgroundAttention and perception are strongly biased toward information about oneself compared to information about others. The self-attention network, an integrative theoretical framework for understanding the self-prioritization effects (SPE), proposes that the ventromedial prefrontal cortex (VMPFC), and the posterior superior temporal sulcus (pSTS) are the two nodes responsible for the preferential processing of self-related stimuli, which interact with the attentional control network (associated with the dorsolateral prefrontal cortex, DLPFC), responsible for processing other-related stimuli. So far, neuroimaging studies have provided considerable correlational evidence supporting the self-attention network.ObjectiveHere we went beyond correlational evidence by manipulating cortical activity using high-definition transcranial direct current stimulation (HD-tDCS), a non-invasive brain stimulation method. We assessed whether anodal and cathodal stimulation of the VMPFC or the DLPFC modulates the processing of self- and other-related stimuli.MethodsWe used an associative unbiased learning procedure, the so-called shape-label matching task, to assess the SPE in a sample of N = 90. We accomplished to overcome different methodological weaknesses of previous studies using different multichannel montages for excitatory and inhibitory effects over both the VMPFC and the DLPFC.ResultsWe found no effect of shape association for non-matching pairs, whereas there was an effect of shape association in the matching condition. Performance (reaction times and accuracy) was better for the self association than for the other two associations, and performance for the friend association was better than for the stranger association. Thus, we replicated the SPE with behavioral data. At the neural level, none of the stimulation succeeded to modulate the magnitude of the SPE effect.ConclusionWe discuss the implications of these findings, in particular why cognitive modeling theories about SPEs should favor an epiphenomenal rather than a causal link between VMPFC/DLPFC and the impact of personal significance stimuli on perception.

Association between Epstein-Barr virus infection and gastric cancer: a systematic review and meta-analysis

BackgroundNumerous studies conducted over the past 30 years have pointed to the presence of Epstein–Barr virus (EBV) in gastric cancer samples. This study was aimed to provide a meta-analytic review of the prevalence of EBV in gastric cancer patients, and to clarify the relationship between EBV infection and gastric cancer.MethodsA literature search was performed electronically using online databases for English language publications until July 1, 2019. The pooled EBV prevalence and 95% confidence intervals (CIs) were estimated using a random-effects model. To determine the association between EBV and gastric cancer, pooled odds ratio (OR) and its 95% CI were computed for case-control studies. Two separate analyses were performed on data from case-control studies with matched and non-match pairs designs to calculate the pooled estimates of ORs.ResultsThe pooled prevalence of EBV in 20,361 gastric cancer patients was 8.77% (95% CI: 7.73–9.92%; I2 = 83.2%). There were 20 studies with matched pairs design, including tumor and tumor-adjacent normal tissue pairs from 4116 gastric cancer patients. The pooled ORs were 18.56 (95% CI: 15.68–21.97; I2 = 55.4%) for studies with matched pairs design and 3.31 (95% CI: 0.95–11.54; I2 = 55.0%) for studies with non-matched pairs design. The proportion of EBV-associated gastric cancer among male cases was significantly higher than among female cases (10.83%, vs. 5.72%) (P < 0.0001). However, the pooled OR estimate for EBV-associated gastric cancer was significantly higher among females (21.47; 95% CI: 15.55–29.63; I2 = 0%) than in males (14.07; 95% CI: 10.46–18.93; I2 = 49.0%) (P = 0.06). EBV was more prevalent in the cardia (12.47%) and the body (11.68%) compared to the antrum (6.29%) (P = 0.0002).ConclusionsEBV infection is associated with more than 18 times increase the risk of gastric cancer. Although the prevalence of EBV was higher in male patients than in female patients with gastric cancer, women are more likely than men to develop EBV-associated gastric cancer. Our findings showed that using tumor-adjacent normal tissues as the control group provides more robust and accurate results regarding the relationship between EBV infection and gastric cancer.

Similarity Filtering with Multibit Trees for Record Linkage

Record linkage is the process of identifying pairs of records that refer to the same real-world object within or across data files. Basically, each record pair is compared with a similarity function and then classified in supposedly matching and non-matching pairs. However, if every possible record pair has to be compared, the resulting number of comparisons leads to infeasible running times for large data files. In such situations, blocking or indexing methods to reduce the comparison space are required. In this paper we propose a new blocking technique (Q-gram Fingerprinting) that efficiently filters record pairs according to an approximation of a q-gram similarity function. The new method first transforms data records into bit vectors, the fingerprints, and then filters pairs of fingerprints by use of a Multibit Tree according to a user-defined similarity threshold. We examined the effect of different parameter choices of Q-gram Fingerprinting, tested its scalability, and performed a comparison study including several alternative methods using simulated person data. The comparison study showed promising results for the proposed method.

A Deep Quadruplet Network for Local Descriptor Learning

Generating a distinguishable feature for local patch is a main task in computer vision which aims at matching local patches. Recently, local patch descriptors from deep convolutional neural network (CNN) with a triplet loss have achieved promising performance. In this paper, we design a quadruplet loss, which can achieve a better result than other pairwise loss and triplet loss methods. Our loss is inspired by the thoughts of uniform distribution. It separates non-matching examples by using the hard sampled non-matching pairs in a batch, and simultaneously uses the random sampled non-matching examples to keep non-matching pairs to obey uniform distribution. A compact descriptor named QuadrupletNet is generated by combining the proposed quadruplet loss and L2Net CNN architecture. From our experiment, QuadrupletNet shows better performance on the Brown dataset and Hpatches dataset than Triplet loss methods on the same training set. The pre-trained QuadrupletNet is publicly available.