Search Results (7)

Eye movement detection algorithms (e.g., I-VT) require the selection of thresholds to identify eye fixations and saccadic movements from gaze data. The choice of threshold is important, as thresholds too low or large may fail to accurately identify eye fixations and saccades. An inaccurate threshold might also affect the resulting visual scan path, the time-ordered sequence of eye fixations and saccades, carried out by the participant. Commonly used approaches to evaluate threshold accuracy can be manually laborious, or require information about the expected visual scan paths of participants, which might not be available. To address this issue, we propose two different computational approaches, labeled as “between-participants comparisons” and “within-participants comparisons.” The approaches were evaluated using the open-source Gazebase dataset, which contained a bullseyetarget tracking task, where participants were instructed to follow the movements of a bullseye-target. The predetermined path of the bullseye-target enabled us to evaluate our proposed approaches against the expected visual scan path. The approaches identified threshold values (220°/s and 210°/s) that were 83% similar to the expected visual scan path, outperforming a 30°/s benchmark threshold (41.5%). These methods might assist researchers in identifying accurate threshold values for the IVT algorithm or potentially other eye movement detection algorithms. Full article

Introduction: Early detection of type 2 diabetes is essential for preventing long-term complications. However, screening the entire population for diabetes is not cost-effective, so identifying individuals at high risk for this disease is crucial. The aim of this study was to compare the performance of five diverse machine learning (ML) models in classifying undiagnosed diabetes using large heterogeneous datasets. Methods: We used machine learning data from several years of the National Health and Nutrition Examination Survey (NHANES) from 2005 to 2018 to identify people with undiagnosed diabetes. The dataset included 45,431 participants, and biochemical confirmation of glucose control (HbA1c) were used to identify undiagnosed diabetes. The predictors were based on simple and clinically obtainable variables, which could be feasible for prescreening for diabetes. We included five ML models for comparison: random forest, AdaBoost, RUSBoost, LogitBoost, and a neural network. Results: The prevalence of undiagnosed diabetes was 4%. For the classification of undiagnosed diabetes, the area under the ROC curve (AUC) values were between 0.776 and 0.806. The positive predictive values (PPVs) were between 0.083 and 0.091, the negative predictive values (NPVs) were between 0.984 and 0.99, and the sensitivities were between 0.742 and 0.871. Conclusion: We have demonstrated that several types of classification models can accurately classify undiagnosed diabetes from simple and clinically obtainable variables. These results suggest that the use of machine learning for prescreening for undiagnosed diabetes could be a useful tool in clinical practice. Full article

Methane (CH₄) hydrate dissociation and CH₄ release are potential geohazards currently investigated using X-ray computed tomography (XCT). Image segmentation is an important data processing step for this type of research. However, it is often time consuming, computing resource-intensive, operator-dependent, and tailored for each XCT dataset due to differences in greyscale contrast. In this paper, an investigation is carried out using U-Nets, a class of Convolutional Neural Network, to segment synchrotron XCT images of CH₄-bearing sand during hydrate formation, and extract porosity and CH₄ gas saturation. Three U-Net deployments previously untried for this task are assessed: (1) a bespoke 3D hierarchical method, (2) a 2D multi-label, multi-axis method and (3) RootPainter, a 2D U-Net application with interactive corrections. U-Nets are trained using small, targeted hand-annotated datasets to reduce operator time. It was found that the segmentation accuracy of all three methods surpass mainstream watershed and thresholding techniques. Accuracy slightly reduces in low-contrast data, which affects volume fraction measurements, but errors are small compared with gravimetric methods. Moreover, U-Net models trained on low-contrast images can be used to segment higher-contrast datasets, without further training. This demonstrates model portability, which can expedite the segmentation of large datasets over short timespans. Full article

There is a high increase in approaches that receive as input a text and perform named entity recognition (or extraction) for linking the recognized entities of the given text to RDF Knowledge Bases (or datasets). In this way, it is feasible to retrieve more information for these entities, which can be of primary importance for several tasks, e.g., for facilitating manual annotation, hyperlink creation, content enrichment, for improving data veracity and others. However, current approaches link the extracted entities to one or few knowledge bases, therefore, it is not feasible to retrieve the URIs and facts of each recognized entity from multiple datasets and to discover the most relevant datasets for one or more extracted entities. For enabling this functionality, we introduce a research prototype, called LODsyndesis${}_{IE}$, which exploits three widely used Named Entity Recognition and Disambiguation tools (i.e., DBpedia Spotlight, WAT and Stanford CoreNLP) for recognizing the entities of a given text. Afterwards, it links these entities to the LODsyndesis knowledge base, which offers data enrichment and discovery services for millions of entities over hundreds of RDF datasets. We introduce all the steps of LODsyndesis${}_{IE}$, and we provide information on how to exploit its services through its online application and its REST API. Concerning the evaluation, we use three evaluation collections of texts: (i) for comparing the effectiveness of combining different Named Entity Recognition tools, (ii) for measuring the gain in terms of enrichment by linking the extracted entities to LODsyndesis instead of using a single or a few RDF datasets and (iii) for evaluating the efficiency of LODsyndesis${}_{IE}$. Full article

Gaze estimation problem can be addressed using either model-based or appearance-based approaches. Model-based approaches rely on features extracted from eye images to fit a 3D eye-ball model to obtain gaze point estimate while appearance-based methods attempt to directly map captured eye images to gaze point without any handcrafted features. Recently, availability of large datasets and novel deep learning techniques made appearance-based methods achieve superior accuracy than model-based approaches. However, many appearance- based gaze estimation systems perform well in within-dataset validation but fail to provide the same degree of accuracy in cross-dataset evaluation. Hence, it is still unclear how well the current state-of-the-art approaches perform in real-time in an interactive setting on unseen users. This paper proposes I2DNet, a novel architecture aimed to improve subject- independent gaze estimation accuracy that achieved a state-of-the-art 4.3 and 8.4 degree mean angle error on the MPIIGaze and RT-Gene datasets respectively. We have evaluated the proposed system as a gaze-controlled interface in real-time for a 9-block pointing and selection task and compared it with Webgazer.js and OpenFace 2.0. We have conducted a user study with 16 participants, and our proposed system reduces selection time and the number of missed selections statistically significantly compared to other two systems. Full article

Distributed monitoring sensor networks are used in an ever increasing number of applications, particularly with the advent of IoT technologies. This has led to a growing demand for unconventional analytical tools to cope with a large amount of different signals. In this scenario, the modeling of time series in similar groups represents an interesting area especially for feature subset selection (FSS) purposes. Methods based on clustering algorithms are very promising for FSS, but in their original form they are unsuitable to manage the complexity of temporal dynamics in time series. In this paper we propose a clustering approach, based on complex network analysis, for the unsupervised FSS of time series in sensor networks. We used natural visibility graphs to map signal segments in the network domain, then extracted features in the form of node degree sequences of the graphs, and finally computed time series clustering through community detection algorithms. The approach was tested on multivariate signals monitored in a 1 MW cogeneration plant and the results show that it outperforms standard time series clustering in terms of both redundancy reduction and information gain. In addition, the proposed method demonstrated its merit in terms of retention of information content with respect to the original dataset in the analyzed condition monitoring system. Full article

The Eye Fixation Related Potential (EFRP) estimation is the average of EEG signals across epochs at ocular fixation onset. Its main limitation is the overlapping issue. Inter Fixation Intervals (IFI) - typically around 300 ms in the case of unrestricted eye movement- depend on participants’ oculomotor patterns, and can be shorter than the latency of the components of the evoked potential. If the duration of an epoch is longer than the IFI value, more than one fixation can occur, and some overlapping between adjacent neural responses ensues. The classical average does not take into account either the presence of several fixations during an epoch or overlapping. The Adjacent Response algorithm (ADJAR), which is popular for event-related potential estimation, was compared to the General Linear Model (GLM) on a real dataset from a conjoint EEG and eye-tracking experiment to address the overlapping issue. The results showed that the ADJAR algorithm was based on assumptions that were too restrictive for EFRP estimation. The General Linear Model appeared to be more robust and efficient. Different configurations of this model were compared to estimate the potential elicited at image onset, as well as EFRP at the beginning of exploration. These configurations took into account the overlap between the event-related potential at stimulus presentation and the following EFRP, and the distinction between the potential elicited by the first fixation onset and subsequent ones. The choice of the General Linear Model configuration was a tradeoff between assumptions about expected behavior and the quality of the EFRP estimation: the number of different potentials estimated by a given model must be controlled to avoid erroneous estimations with large variances. Full article