Search

Review studies play a key role in the development of clinical practice by synthesizing data and drawing conclusions from multiple scientific sources. In recent decades, there has been a significant increase in the number of review studies conducted and published by researchers. In clinical research, different types of review studies (systematic, scoping, umbrella, and narrative reviews) are conducted with different objectives and methodologies. Despite the abundance of guidelines for conducting review studies, researchers often face challenges in selecting the most appropriate review method, mainly due to their overlapping characteristics, including the complexity of matching review types to specific research questions. The aim of this article is to compare the main features of systematic, scoping, umbrella, and narrative reviews in clinical research and to address key considerations for selecting the most appropriate review approach. It also discusses future opportunities for updating their strategies based on emerging trends in clinical research. Understanding the differences between review approaches will help researchers, practitioners, journalists, and policymakers to effectively navigate the complex and evolving field of scientific research, leading to informed decisions that ultimately enhance the overall quality of healthcare practices. [ABSTRACT FROM AUTHOR]

Chemical exchange saturation transfer (CEST) MRI at 3 T suffers from low specificity due to overlapping CEST effects from multiple metabolites, while higher field strengths (B0) allow for better separation of Z‐spectral "peaks," aiding signal interpretation and quantification. However, data acquisition at higher B0 is restricted by equipment access, field inhomogeneity and safety issues. Herein, we aim to synthesize higher‐B0Z‐spectra from readily available data acquired with 3 T clinical scanners using a deep learning framework. Trained with simulation data using models based on Bloch–McConnell equations, this framework comprised two deep neural networks (DNNs) and a singular value decomposition (SVD) module. The first DNN identified B0 shifts in Z‐spectra and aligned them to correct frequencies. After B0 correction, the lower‐B0Z‐spectra were streamlined to the second DNN, casting into the key feature representations of higher‐B0Z‐spectra, obtained through SVD truncation. Finally, the complete higher‐B0Z‐spectra were recovered from inverse SVD, given the low‐rank property of Z‐spectra. This study constructed and validated two models, a phosphocreatine (PCr) model and a pseudo‐in‐vivo one. Each experimental dataset, including PCr phantoms, egg white phantoms, and in vivo rat brains, was sequentially acquired on a 3 T human and a 9.4 T animal scanner. Results demonstrated that the synthetic 9.4 T Z‐spectra were almost identical to the experimental ground truth, showing low RMSE (0.11% ± 0.0013% for seven PCr tubes, 1.8% ± 0.2% for three egg white tubes, and 0.79% ± 0.54% for three rat slices) and high R2 (>0.99). The synthesized amide and NOE contrast maps, calculated using the Lorentzian difference, were also well matched with the experiments. Additionally, the synthesis model exhibited robustness to B0 inhomogeneities, noise, and other acquisition imperfections. In conclusion, the proposed framework enables synthesis of higher‐B0Z‐spectra from lower‐B0 ones, which may facilitate CEST MRI quantification and applications. [ABSTRACT FROM AUTHOR]

Background: The potential of ursodeoxycholic acid (UDCA) in inhibiting angiotensin-converting enzyme 2 was demonstrated. However, conflicting evidence emerged regarding the association between UDCA and COVID-19 outcomes, prompting the need for a comprehensive investigation. Research design and methods: Patients diagnosed with COVID-19 infection were retrospectively analyzed and divided into two groups: the UDCA-treated group and the control group. Kaplan–Meier recovery analysis and Cox proportional hazards models were used to evaluate the recovery time and hazard ratios. Additionally, study-level pooled analyses for multiple clinical outcomes were performed. Results: In the 115-patient cohort, UDCA treatment was significantly associated with a reduced recovery time. The subgroup analysis suggests that the 300 mg subgroup had a significant (adjusted hazard ratio: 1.63 [95% CI, 1.01 to 2.60]) benefit with a shorter duration of fever. The results of pooled analyses also show that UDCA treatment can significantly reduce the incidence of severe/critical diseases in COVID-19 (adjusted odds ratio: 0.68 [95% CI, 0.50 to 0.94]). Conclusions: UDCA treatment notably improves the recovery time following an Omicron strain infection without observed safety concerns. These promising results advocate for UDCA as a viable treatment for COVID-19, paving the way for further large-scale and prospective research to explore the full potential of UDCA. [ABSTRACT FROM AUTHOR]

Coronary Artery Bypass Graft (CABG) is a common cardiac surgery, but it continues to have many associated risks, including the need for blood transfusions. Previous research has shown that blood transfusion during CABG surgery is associated with an increased risk for infection and mortality. The current study aims to use modern techniques, such as deep neural networks and data synthesis, to develop models that can best predict the need for blood transfusion among CABG patients. Results show that neural networks with synthetic data generated by DataSynthesizer have the best performance. Implications of results and future directions are discussed. [ABSTRACT FROM AUTHOR]

Utilising novel ways of knowing, aligned with an ecological approach, the Learning in Development Research Framework (LDRF) has been introduced as a different way to guide research and practice in sport. A central feature of this framework is an appreciation of researcher embeddedness; positioned as an inhabitant who follows along with the unfolding inquiry. This positioning is integral for enriching ones understanding of the relations between socio-cultural constraints and affordances for skill learning within a sports organisation. Moreover, the notion of embeddedness foregrounds the ongoing nature of inquiry when practiced as an art of inquiry. In an effort to extend these ideas, this paper highlights how a phronetic iterative approach to data analysis-synthesis could be undertaken, while ensuring that the researcher remains 'in touch' with a phenomenon, and thus faithful to key tenets of research practiced as an art of inquiry. To illustrate this, we present a 'walk-through' from a recent LDRF study. Rather than focusing on data collection or recorded observations made from afar, this walk-through shows how a researcher, practicing an art of inquiry, can grow knowledge of and with the phenomena, enriching the evolution of practice and performance from within an ecology of relations. [ABSTRACT FROM AUTHOR]

Copyright of Systems Engineering & Electronics is the property of Journal of Systems Engineering & Electronics Editorial Department and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Time series forecasting has witnessed a rapid proliferation of novel neural network approaches in recent times. However, performances in terms of benchmarking results are generally not consistent, and it is complicated to determine in which cases one approach fits better than another. Therefore, we propose adopting a data-centric perspective for benchmarking neural network architectures on time series forecasting by generating ad hoc synthetic datasets. In particular, we combine sinusoidal functions to synthesize univariate time series data for multi-input-multi-output prediction tasks. We compare the most popular architectures for time series, namely long short-term memory (LSTM) networks, convolutional neural networks (CNNs), and transformers, and directly connect their performance with different controlled data characteristics, such as the sequence length, noise and frequency, and delay length. Our findings suggest that transformers are the best architecture for dealing with different delay lengths. In contrast, for different noise and frequency levels and different sequence lengths, LSTM is the best-performing architecture by a significant amount. Based on our insights, we derive recommendations which allow machine learning (ML) practitioners to decide which architecture to apply, given the dataset’s characteristics.

In an era of global environmental change, empirical synthesis of biodiversity data across geographic scales and taxonomic groups is urgently required. Recently, with an upsurge in data synthesis, substantial progress has been made in making massive biodiversity data available on a global scale. However, most of these databases lack sufficient geographic coverage, particularly from biodiversity hotspot regions of developing countries. Here, we present a comprehensive and curated plant database of the Indian Himalayan Region (IHR) – home to two global biodiversity hotspots. The database, currently comprising 11,743 native plant species, has been collated from an extensive quantitative synthesis of 324 floristic studies published between 1872 and 2022. Based on this database, we investigate the patterns of species richness, distribution, life-history traits, endemic and threat status of the native flora of the IHR, and the results revealed that these patterns vary considerably among the 12 states of the IHR. Sikkim harbours the highest number of plant species (5090), followed by Arunachal Pradesh (4907). We found a total of 1123 species (ca. 10%) endemic to India and 157 threatened species occurring in the IHR. The life-history traits (growth form and lifespan) were unequally represented between the Himalaya and the Indo-Burma hotspots. We found herbs as the dominant growth form across the IHR. Also, maximum species similarity was found between Jammu and Kashmir and Himachal Pradesh (Cs = 0.66), and minimum between the former and Meghalaya (Cs = 0.13). Overall, our study represents a significant step forward in filling the knowledge gaps from the global biodiversity hotspots in India, with immense management and policy implications. [ABSTRACT FROM AUTHOR]

Time series forecasting has witnessed a rapid proliferation of novel neural network approaches in recent times. However, performances in terms of benchmarking results are generally not consistent, and it is complicated to determine in which cases one approach fits better than another. Therefore, we propose adopting a data-centric perspective for benchmarking neural network architectures on time series forecasting by generating ad hoc synthetic datasets. In particular, we combine sinusoidal functions to synthesize univariate time series data for multi-input-multi-output prediction tasks. We compare the most popular architectures for time series, namely long short-term memory (LSTM) networks, convolutional neural networks (CNNs), and transformers, and directly connect their performance with different controlled data characteristics, such as the sequence length, noise and frequency, and delay length. Our findings suggest that transformers are the best architecture for dealing with different delay lengths. In contrast, for different noise and frequency levels and different sequence lengths, LSTM is the best-performing architecture by a significant amount. Based on our insights, we derive recommendations which allow machine learning (ML) practitioners to decide which architecture to apply, given the dataset's characteristics. [ABSTRACT FROM AUTHOR]

Chlorophyll-a (Chl-a) is a pivotal indicator of lake eutrophication. Studies examining nutrients limiting lake eutrophication at large scales have traditionally focused on summer and autumn, potentially limiting the applicability of their findings. This study encompasses 86 state-controlled points in the Eastern China Basin, spanning data collected from January 2020 to July 2023. Furthermore, we focus on the application of three machine-learning models (i.e., eXtreme Gradient Boosting, Support Vector Machines, and Naive Bayes Classifier) to analyze the seasonal nutrient dynamics in lake ecosystems. We categorized the monitoring data by season to eliminate outliers and employed adaptive synthetic sampling to address data imbalance issues. The results reveal that the direct correlations between total nitrogen (TN), total phosphorus (TP), and TP in conjunction with turbidity and Chl-a are broadly weak, possibly because of geographic variations, nutrient lag effects on algae, and differences in algal community composition. However, probabilistic analyses revealed that as TP or TN levels transitioned from oligo-mesotrophic (O) to eutrophic (E), TP exhibited a greater influence on the variation in Chl-a status than TN during spring and winter (p

Abstract: The San Francisco Estuary, in central California, has several long-running monitoring programs that have been used to reveal human-induced changes throughout the estuary in the last century. Here, we pair synoptic records of particulate organic matter (POM) composition from 1990–1996 and 2007–2016 with more robust long-term monitoring program records of total suspended sediment (TSS) concentrations generally starting in the mid-1970s to better understand how POM and TSS sources and transport have shifted. Specifically, POM C:N ratios and stable isotope values were used as indicators of POM source and to separate the bulk POC pool into detrital and phytoplankton components. We found that TSS and POC sources have shifted significantly across the estuary in time and space from declines in terrestrial inputs. Landward freshwater and brackish water sites, in the Delta and near Suisun Bay, witnessed long-term declines in TSS (32 to 52%), while seaward sites, near San Pablo Bay, recorded recent increases in TSS (16 to 121%) that began to trend downwards at the end of the record considered. Bulk POM C:N ratios shifted coeval with the TSS concentration changes at nearly all sites, with mean declines of 12 to 27% between 1990–1996 and 2007–2016. The widespread declines in bulk POM C:N ratios and inferred changes in POC concentrations from TSS trends, along with the substantial declines in upstream TSS supply through time (56%), suggest measurable reductions in terrestrial inputs to the system. Changes in terrestrial TSS and POM inputs have implications for biotic (e.g., food web dynamics) and abiotic organic matter cycling (e.g., burial, export) along the estuarine continuum. This work demonstrates how human-generated environmental changes can propagate spatially and temporally through a large river-estuary system. More broadly, we show how underutilized monitoring program datasets can be paired with existing (and often imperfect) synoptic records to generate new system insight in lieu of new data collection.

Cardiovascular diseases remain one of the main threats to human health, significantly affecting the quality and life expectancy. Effective and prompt recognition of these diseases is crucial. This research aims to develop an effective novel hybrid method for automatically detecting dangerous arrhythmias based on cardiac patients’ short electrocardiogram (ECG) fragments. This study suggests using a continuous wavelet transform (CWT) to convert ECG signals into images (scalograms) and examining the task of categorizing short 2-s segments of ECG signals into four groups of dangerous arrhythmias that are shockable, including ventricular flutter (C1), ventricular fibrillation (C2), ventricular tachycardia torsade de pointes (C3), and high-rate ventricular tachycardia (C4). We propose developing a novel hybrid neural network with a deep learning architecture to classify dangerous arrhythmias. This work utilizes actual electrocardiogram (ECG) data obtained from the PhysioNet database, alongside artificially generated ECG data produced by the Synthetic Minority Over-sampling Technique (SMOTE) approach, to address the issue of imbalanced class distribution for obtaining an accuracy-trained model. Experimental results demonstrate that the proposed approach achieves high accuracy, sensitivity, specificity, precision, and an F1-score of 97.75%, 97.75%, 99.25%, 97.75%, and 97.75%, respectively, in classifying all the four shockable classes of arrhythmias and are superior to traditional methods. Our work possesses significant clinical value in real-life scenarios since it has the potential to significantly enhance the diagnosis and treatment of lifethreatening arrhythmias in individuals with cardiac disease. Furthermore, our model also has demonstrated adaptability and generality for two other datasets. [ABSTRACT FROM AUTHOR]

Data pooling is a powerful strategy in empirical research. However, combining multiple datasets often results in a large amount of missing data, as variables that are not present in some datasets effectively contain missing values for all participants in those datasets. Furthermore, data pooling typically leads to a mix of continuous and categorical items with nonnormal multivariate distributions. We investigated two popular approaches to handle missing data in this context: (1) applying direct maximum likelihood by treating data as continuous (con-ML), and (2) applying categorical least squares using a polychoric correlation matrix computed from pairwise deletion (cat-LS). These approaches are available for free and relatively straightforward for empirical researchers to implement. Through simulation studies with confirmatory factor analysis and latent mediation analysis, we found cat-LS to be unsuitable for pooled data analysis, whereas con-ML yielded acceptable performance for the estimation of latent path coefficients barring severe nonnormality. [ABSTRACT FROM AUTHOR]

Deep learning models have been used for a variety of image processing tasks. However, most of these models are developed through supervised learning approaches, which rely heavily on the availability of large-scale annotated datasets. Developing such datasets is tedious and expensive. In the absence of an annotated dataset, synthetic data can be used for model development; however, due to the substantial differences between simulated and real data, a phenomenon referred to as domain gap, the resulting models often underperform when applied to real data. In this research, we aim to address this challenge by first computationally simulating a large-scale annotated dataset and then using a generative adversarial network (GAN) to fill the gap between simulated and real images. This approach results in a synthetic dataset that can be effectively utilized to train a deep-learning model. Using this approach, we developed a realistic annotated synthetic dataset for wheat head segmentation. This dataset was then used to develop a deep-learning model for semantic segmentation. The resulting model achieved a Dice score of 83.4% on an internal dataset and Dice scores of 79.6% and 83.6% on two external datasets from the Global Wheat Head Detection datasets. While we proposed this approach in the context of wheat head segmentation, it can be generalized to other crop types or, more broadly, to images with dense, repeated patterns such as those found in cellular imagery. [ABSTRACT FROM AUTHOR]

Heterogeneous data have been used to enhance behavior prediction performance; however, it involves issues such as missing data, which need to be addressed. This paper proposes enhanced pet behavior prediction via Sensor to Skeleton Generative Adversarial Networks (S2GAN)-based heterogeneous data synthesis. The S2GAN model synthesizes the key features of video skeletons based on collected nine-axis sensor data and replaces missing data, thereby enhancing the accuracy of behavior prediction. In this study, data collected from 10 pets in a real-life-like environment were used to conduct recognition experiments on 9 commonly occurring types of indoor behavior. Experimental results confirmed that the proposed S2GAN-based synthesis method effectively resolves possible missing data issues in real environments and significantly improves the performance of the pet behavior prediction model. Additionally, by utilizing data collected under conditions similar to the real environment, the method enables more accurate and reliable behavior prediction. This research demonstrates the importance and utility of synthesizing heterogeneous data in behavior prediction, laying the groundwork for applications in various fields such as abnormal behavior detection and monitoring. [ABSTRACT FROM AUTHOR]

Abstract Background This study examined the synthesis methods used in meta-analyses pooling data from observational studies (OSs) and randomised controlled trials (RCTs) from various medical disciplines. Methods We searched Medline via PubMed to identify reports of systematic reviews of interventions, including and pooling data from RCTs and OSs published in 110 high-impact factor general and specialised journals between 2015 and 2019. Screening and data extraction were performed in duplicate. To describe the synthesis methods used in the meta-analyses, we considered the first meta-analysis presented in each article. Results Overall, 132 reports were identified with a median number of included studies of 14 [9–26]. The median number of OSs was 6.5 [3–12] and that of RCTs was 3 [1–6]. The effect estimates recorded from OSs (i.e., adjusted or unadjusted) were not specified in 82% (n = 108) of the meta-analyses. An inverse-variance common-effect model was used in 2% (n = 3) of the meta-analyses, a random-effects model was used in 55% (n = 73), and both models were used in 40% (n = 53). A Poisson regression model was used in 1 meta-analysis, and 2 meta-analyses did not report the model they used. The mean total weight of OSs in the studied meta-analyses was 57.3% (standard deviation, ± 30.3%). Only 44 (33%) meta-analyses reported results stratified by study design. Of them, the results between OSs and RCTs had a consistent direction of effect in 70% (n = 31). Study design was explored as a potential source of heterogeneity in 79% of the meta-analyses, and confounding factors were investigated in only 10% (n = 13). Publication bias was assessed in 70% (n = 92) of the meta-analyses. Tau-square was reported in 32 meta-analyses with a median of 0.07 [0–0.30]. Conclusion The inclusion of OSs in a meta-analysis on interventions could provide useful information. However, considerations of several methodological and conceptual aspects of OSs, that are required to avoid misleading findings, were often absent or insufficiently reported in our sample.

Due to the increasing severity of network security issues, training corresponding detection models requires large datasets. In this work, we propose a novel method based on generative adversarial networks to synthesize network data traffic. We introduced a network traffic data normalization method based on Gaussian mixture models (GMM), and for the first time, incorporated a generator based on the Swin Transformer structure into the field of network traffic generation. To further enhance the robustness of the model, we mapped real data through an AE (autoencoder) module and optimized the training results in the form of evolutionary algorithms. We validated the training results on four different datasets and introduced four additional models for comparative experiments in the experimental evaluation section. Our proposed SEGAN outperformed other state-of-the-art network traffic emulation methods. [ABSTRACT FROM AUTHOR]

Cardiovascular diseases remain one of the main threats to human health, significantly affecting the quality and life expectancy. Effective and prompt recognition of these diseases is crucial. This research aims to develop an effective novel hybrid method for automatically detecting dangerous arrhythmias based on cardiac patients’ short electrocardiogram (ECG) fragments. This study suggests using a continuous wavelet transform (CWT) to convert ECG signals into images (scalograms) and examining the task of categorizing short 2-s segments of ECG signals into four groups of dangerous arrhythmias that are shockable, including ventricular flutter (C1), ventricular fibrillation (C2), ventricular tachycardia torsade de pointes (C3), and high-rate ventricular tachycardia (C4). We propose developing a novel hybrid neural network with a deep learning architecture to classify dangerous arrhythmias. This work utilizes actual electrocardiogram (ECG) data obtained from the PhysioNet database, alongside artificially generated ECG data produced by the Synthetic Minority Over-sampling Technique (SMOTE) approach, to address the issue of imbalanced class distribution for obtaining an accuracy-trained model. Experimental results demonstrate that the proposed approach achieves high accuracy, sensitivity, specificity, precision, and an F1-score of 97.75%, 97.75%, 99.25%, 97.75%, and 97.75%, respectively, in classifying all the four shockable classes of arrhythmias and are superior to traditional methods. Our work possesses significant clinical value in real-life scenarios since it has the potential to significantly enhance the diagnosis and treatment of life-threatening arrhythmias in individuals with cardiac disease. Furthermore, our model also has demonstrated adaptability and generality for two other datasets.

Data augmentation involves artificially expanding a dataset by applying various transformations to the existing data. Recent developments in deep learning have advanced data augmentation, enabling more complex transformations. Especially vital in the medical domain, deep learning-based data augmentation improves model robustness by generating realistic variations in medical images, enhancing diagnostic and predictive task performance. Therefore, to assist researchers and experts in their pursuits, there is a need for an extensive and informative study that covers the latest advancements in the growing domain of deep learning-based data augmentation in medical imaging. There is a gap in the literature regarding recent advancements in deep learning-based data augmentation. This study explores the diverse applications of data augmentation in medical imaging and analyzes recent research in these areas to address this gap. The study also explores popular datasets and evaluation metrics to improve understanding. Subsequently, the study provides a short discussion of conventional data augmentation techniques along with a detailed discussion on applying deep learning algorithms in data augmentation. The study further analyzes the results and experimental details from recent state-of-the-art research to understand the advancements and progress of deep learning-based data augmentation in medical imaging. Finally, the study discusses various challenges and proposes future research directions to address these concerns. This systematic review offers a thorough overview of deep learning-based data augmentation in medical imaging, covering application domains, models, results analysis, challenges, and research directions. It provides a valuable resource for multidisciplinary studies and researchers making decisions based on recent analytics.

With the vigorously developed services of cloud computing, it is relatively easier and more convenient for organizations or enterprises to open data on clouds. However, as personal information in electronic data becomes more massive and detailed, how to balance data opening and personal privacy has become a critical issue. In this paper, we propose the Theseus Data Synthesis Approach (TDSA), which generates synthetic data by replacing partial records until no record from the original dataset remains. Unlike other data anonymization works such as k-anonymity and differential privacy, which encountered limitations and challenges when applying to real-world scenarios. In our work, Since there are no real data, personal privacy is definitely preserved. We also analyze the quality and utility of the synthetic dataset and make comparisons with related works. We conclude that with our scheme, opening useful data on clouds and preserving personal privacy can be simultaneously achieved.

Classification models using deep or machine learning algorithms require a sufficient and balanced training dataset to improve performance. Still, they suffer from data collection due to data privacy issues. In medical research, where most data variables are sensitive information, collecting enough training data for model performance improvement is more challenging. This study presents a new medical data augmentation algorithm consisting of four steps to solve the data shortage and class imbalance issues. The main idea of the proposed algorithm is to reflect the core characteristic of the original data’s class label. The algorithm receives an original dataset as an input value to extract the feature vector and trains the individual autoencoder model. Then it verifies the augmented feature vector through a distributional equality check, and each feature vector is concatenated into one feature vector. The deep learning model inference is applied on a concatenated vector for the second verification, to finalize the augmented training dataset. Our team performed mood disorder classification using patient data to prove the presented data augmentation algorithm. With the method, the classification performance improved by 0.059 in the severity classification of major depressive disorder, 0.041 in the severity classification of anxiety disorder, and 0.073 in the subtype classification of bipolar disorder. Through this study, we proved that our algorithm can be applied to minimize model bias and improve classification performance on the medical data that are unbalanced or insufficient in number by class.

Abstract As the number one building material, concrete is of fundamental importance in civil engineering. Understanding its failure mechanisms is essential for designing sustainable buildings and infrastructure. Micro-computed tomography (μCT) is a well-established tool for virtually assessing crack initiation and propagation in concrete. The reconstructed 3d images can be examined via techniques from the fields of classical image processing and machine learning. Ground truths are a prerequisite for an objective evaluation of crack segmentation methods. Furthermore, they are necessary for training machine learning models. However, manual annotation of large 3d concrete images is not feasible. To tackle the problem of data scarcity, the image pairs of cracked concrete and corresponding ground truth can be synthesized. In this work we propose a novel approach to stochastically model crack structures via Voronoi diagrams. The method is based on minimum-weight surfaces, an extension of shortest paths to 3d. Within a dedicated image processing pipeline, the surfaces are then discretized and embedded into real μCT images of concrete. The method is flexible and fast, such that a variety of different crack structures can be generated in a short amount of time.

Supervised Neural Networks are used for segmentation in many biological and biomedical applications. To omit the time-consuming and tiring process of manual labeling, unsupervised Generative Adversarial Networks (GANs) can be used to synthesize labeled data. However, the training of GANs requires extensive computation and is often unstable. Due to the lack of established stopping criteria, GANs are usually trained multiple times for a heuristically fixed number of epochs. Early stopping and epoch selection can lead to better synthetic datasets resulting in higher downstream segmentation quality on biological or medical data. This article examines whether the Frechet Inception Distance (FID), the Kernel Inception Distance (KID), or the WeightWatcher tool can be used for early stopping or epoch selection of unsupervised GANs. The experiments show that the last trained GAN epoch is not necessarily the best one to synthesize downstream segmentation data. On complex datasets, FID and KID correlate with the downstream segmentation quality, and both can be used for epoch selection.

Classification of sheep behaviour from a sequence of tri-axial accelerometer data has the potential to enhance sheep management. Sheep behaviour is inherently imbalanced (e.g., more ruminating than walking) resulting in underperforming classification for the minority activities which hold importance. Existing works have not addressed class imbalance and use traditional machine learning techniques, e.g., Random Forest (RF). We investigated Deep Learning (DL) models, namely, Long Short Term Memory (LSTM) and Bidirectional LSTM (BLSTM), appropriate for sequential data, from imbalanced data. Two data sets were collected in normal grazing conditions using jaw-mounted and ear-mounted sensors. Novel to this study, alongside typical single classes, e.g., walking, depending on the behaviours, data samples were labelled with compound classes, e.g., walking_grazing. The number of steps a sheep performed in the observed 10 s time window was also recorded and incorporated in the models. We designed several multi-class classification studies with imbalance being addressed using synthetic data. DL models achieved superior performance to traditional ML models, especially with augmented data (e.g., 4-Class + Steps: LSTM 88.0%, RF 82.5%). DL methods showed superior generalisability on unseen sheep (i.e., F1-score: BLSTM 0.84, LSTM 0.83, RF 0.65). LSTM, BLSTM and RF achieved sub-millisecond average inference time, making them suitable for real-time applications. The results demonstrate the effectiveness of DL models for sheep behaviour classification in grazing conditions. The results also demonstrate the DL techniques can generalise across different sheep. The study presents a strong foundation of the development of such models for real-time animal monitoring.

Mixed methods research requires data integration from multiple sources. Existing techniques are restricted to integrating a maximum of two data sources, do not provide step-by-step guidance or can be cumbersome where many data need to be integrated. We have solved these limitations through the development of the extended Pillar Integration Process (ePIP), a method which contributes to the field of mixed methods by being the first data integration method providing explicit steps on how to integrate data from three data sources. The ePIP provides greater transparency, validity and consistency compared to existing methods. We provide two worked examples from health sciences and automotive human factors, highlighting its value as a mixed methods integration tool. [ABSTRACT FROM AUTHOR]

When classifiers face imbalanced class distributions, they often misclassify minority class samples, consequently diminishing the predictive performance of machine learning models. Existing oversampling techniques predominantly rely on the selection of neighboring data via interpolation, with less emphasis on uncovering the intrinsic patterns and relationships within the data. In this research, we present the usefulness of an algorithm named RuLer to deal with the problem of classification with imbalanced data. RuLer is a learning algorithm initially designed to recognize new sound patterns within the context of the performative artistic practice known as live coding. This paper demonstrates that this algorithm, once adapted (Ad-RuLer), has great potential to address the problem of oversampling imbalanced data. An extensive comparison with other mainstream oversampling algorithms (SMOTE, ADASYN, Tomek-links, Borderline-SMOTE, and KmeansSMOTE), using different classifiers (logistic regression, random forest, and XGBoost) is performed on several real-world datasets with different degrees of data imbalance. The experiment results indicate that Ad-RuLer serves as an effective oversampling technique with extensive applicability. [ABSTRACT FROM AUTHOR]

As the number one building material, concrete is of fundamental importance in civil engineering. Understanding its failure mechanisms is essential for designing sustainable buildings and infrastructure. Micro-computed tomography (μCT) is a well-established tool for virtually assessing crack initiation and propagation in concrete. The reconstructed 3d images can be examined via techniques from the fields of classical image processing and machine learning. Ground truths are a prerequisite for an objective evaluation of crack segmentation methods. Furthermore, they are necessary for training machine learning models. However, manual annotation of large 3d concrete images is not feasible. To tackle the problem of data scarcity, the image pairs of cracked concrete and corresponding ground truth can be synthesized. In this work we propose a novel approach to stochastically model crack structures via Voronoi diagrams. The method is based on minimum-weight surfaces, an extension of shortest paths to 3d. Within a dedicated image processing pipeline, the surfaces are then discretized and embedded into real μCT images of concrete. The method is flexible and fast, such that a variety of different crack structures can be generated in a short amount of time. [ABSTRACT FROM AUTHOR]

Recently, deep learning (DL) models have been increasingly adopted for automatic analyses of medical data, including electrocardiograms (ECGs). Large, available ECG datasets, generally of high quality, often lack specific distortions, which could be helpful for enhancing DL-based algorithms. Synthetic ECG datasets could overcome this limitation. A generative adversarial network (GAN) was used to synthesize realistic 3D magnetohydrodynamic (MHD) distortion templates, as observed during magnetic resonance imaging (MRI), and then added to available ECG recordings to produce an augmented dataset. Similarity metrics, as well as the accuracy of a DL-based R-peak detector trained with and without data augmentation, were used to evaluate the effectiveness of the synthesized data. Three-dimensional MHD distortions produced by the proposed GAN were similar to the measured ones used as input. The precision of a DL-based R-peak detector, tested on actual unseen data, was significantly enhanced by data augmentation; its recall was higher when trained with augmented data. Using synthesized MHD-distorted ECGs significantly improves the accuracy of a DL-based R-peak detector, with a good generalization capacity. This provides a simple and effective alternative to collecting new patient data. DL-based algorithms for ECG analyses can suffer from bias or gaps in training datasets. Using a GAN to synthesize new data, as well as metrics to evaluate its performance, can overcome the scarcity issue of data availability. [ABSTRACT FROM AUTHOR]

The problem of new image synthesis with the use Neural Radiance Fields (NeRF) for an environment with dynamic illumination is considered. When training NeRF models, a photometric loss function is used, i.e., a pixel-by-pixel difference between intensity values of scene images and the images generated using NeRF. For reflective surfaces, image intensity depends on the viewing angle, and this effect is accounted for by using the direction of a ray as the NeRF model input parameter. For scenes with dynamic illumination, image intensity depends not only on the position and viewing direction, but also on time. It is shown that illumination change affects the learning of NeRF with a standard photometric loss function and decreases the quality of the obtained images and depth maps. To overcome this problem, we propose to introduce time as an additional NeRF input argument. Experiments performed on the ScanNet dataset demonstrate that NeRF with a modified input outperform the original model version and generate more consistent and coherent 3D structures. The results of this study can be used to improve the quality of training data augmentation for training distance forecasting models (e.g., depth-from-stereo models allowing for depth/distance forecasts based on stereo data) for scenes with non-static illumination. [ABSTRACT FROM AUTHOR]