Event Detection Research Articles

AUDIOSIGNAL BASED EVENT DETECTION USING DEEP LEARNING TECHNIQUES

Deep learning has garnered significant interest from researchers for performing pattern recognition tasks. In particular, the detection of events based on audio signals and the recognition of natural sounds in the environment stand out. The DCASE challenge – Detection and Classification of Acoustic Scenes and Events – has further highlighted the efficiency of deep learning in accomplishing these tasks. This paper reviews the works of other researchers that applied various deep learning techniques to detect emergency events based on audio signals. It focuses on the complexity and specific challenges of recognizing polyphonic sound-based events. The use and structures of neural networks are presented, with an emphasis on the application of CNN and RNN for event detection based on audio signals. Evaluation metrics and an overview of datasets are also provided.

A Human-in-the-Loop Method for Annotation of Events in Biomedical Signals.

Building large-scale data bases of biomedical signal recordings for training artificial-intelligence systems involves substantial human effort in data processing and annotation. In the case of event detection, experts need to exhaustively scroll through the recordings and highlight events of interest. We propose an iterative annotation support algorithm with a human in the loop to improve the efficiency of the annotation process. Our algorithm generates proposal events based on an event detection model trained on incomplete annotations. The human only needs to verify candidate events proposed by the tool instead of scrolling through the entire data set. Our algorithm iterates between proposal generation and verification to leverage the human-in-the-loop feedback to obtain a growing set of event annotations. Our algorithm finds a substantial amount of events at a fraction of the human time spent when comparing with a benchmark method and the normal manual process, finding all events in one data set and 70% of events in another with the human-in-the-loop only viewing 20% of the data. Our results show that combining human and computer effort can substantially speed up the annotation process for events in biomedical signal processing. Due to its simplicity and minimal reliance on task-specific information, our algorithm is broadly applicable, unlocking substantial improvements in the scalability and efficiency of biomedical signal annotation.

Relational Prompt-based Pre-trained Language Models for Social Event Detection

Relational Prompt-based Pre-trained Language Models for Social Event Detection

Regulating learning module for patient monitoring interactive event detecting robots

Process automation robots in smart healthcare services are reliable in providing interactive patient services. The patient requirements are identified and approached through automated learning and responses. A Regulating Learning Interaction Module (RLIM) using Sigmoid Optimized Neural Networks is designed for robots in such sensitive environments. This learning module trains the robots for various services/ user interactions over patient demands. In particular, the patient interacting features on monitoring and responses/ event triggering are induced through the learning. The miscommunicated interactions are streamlined using sigmoid optimization from the conditional neural layers for assessment. This assessment delivers the possible choices to improve event detection/ response delivery. The neural network regulates the training patterns from the user inputs, histories, and patient diagnosis data for which the events are detected. This event relates the patient with the diagnosis, assistance, and monitoring for better assistance. The performance of this proposed module is validated using interaction detection, response precision, event generation rate, and miscommunication.

Fast Comparative Analysis of Merge Trees Using Locality Sensitive Hashing.

Scalar field comparison is a fundamental task in scientific visualization. In topological data analysis, we compare topological descriptors of scalar fields-such as persistence diagrams and merge trees-because they provide succinct and robust abstract representations. Several similarity measures for topological descriptors seem to be both asymptotically and practically efficient with polynomial time algorithms, but they do not scale well when handling large-scale, time-varying scientific data and ensembles. In this paper, we propose a new framework to facilitate the comparative analysis of merge trees, inspired by tools from locality sensitive hashing (LSH). LSH hashes similar objects into the same hash buckets with high probability. We propose two new similarity measures for merge trees that can be computed via LSH, using new extensions to Recursive MinHash and subpath signature, respectively. Our similarity measures are extremely efficient to compute and closely resemble the results of existing measures such as merge tree edit distance or geometric interleaving distance. Our experiments demonstrate the utility of our LSH framework in applications such as shape matching, clustering, key event detection, and ensemble summarization.

Skip-AttSeqNet: Leveraging skip connection and attention-driven Seq2seq model to enhance eye movement event detection in Parkinson’s disease

To address the limitations of traditional algorithms in detecting eye movement events, particularly in Parkinson’s disease (PD) patients, this study introduces Skip-AttSeqNet. It presents an innovative approach combining skip-connected, one-dimensional convolutional neural networks with an attention-enhanced, bidirectional long short-term memory network. This hybrid architecture significantly advances smooth pursuit (SP) event detection, as evidenced by its performance on both the GazeCom dataset and a unique dataset of PD patient eye movements. Key innovations in this work include the utilization of skip connections and attention mechanisms, along with optimized training–validation set division, collectively enhancing the model’s accuracy while mitigating overfitting. Skip-AttSeqNet outperforms existing algorithms, achieving a 3.2% higher sample-level F1 score and a notable 6.2% increase in event-level F1 scores for SP detection. Furthermore, we established a smooth-pursuit experimental paradigm and identified significant differences in saccade and SP features between PD patients and healthy older adults through statistical analysis using the Mann–Whitney test. These findings underscore the potential of eye movement metrics as biomarkers for PD, thereby not only strengthening PD diagnosis but also enriching the intersection of computer vision and biomedical research domains.

Prevention, detection and management of adverse events of third-space endoscopy.

Third space endoscopy (TSE) or sub-mucosal endoscopy using a mucosal flap valve (SEMF) enables the endoscopist to operate in the deeper layers of the gastrointestinal tract or gain access to the mediastinal/peritoneal cavity for natural orifice transoral endoscopic surgery (NOTES). TSE procedures are essentially endoscopic surgical procedures with a variable learning curve. Adverse events (AEs) during TSE are specific and follow a certain pattern across the spectrum of TSE procedures. These can be broadly categorized according to either type of AE, time of presentation relative to the procedure or according to degree of severity. Three major categories of AEs encountered during TSE include insufflation related AEs, mucosal injuries (MIs) and bleeding. Other relevant AEs include infectious complications, aspiration pneumonia, post-procedural chest/abdominal pain, atelectasis, cardiac arrhythmias, pleural effusion and pulmonary embolism. Reported incidence of AEs during TSE procedures varies according to the type and complexity of procedure. Acquaintance regarding potential risk factors, technical tips and precautions, alarm signs for early recognition, assessment of degree of severity, morphological characterization of AEs and finally, expeditious selection of appropriate management strategy are crucial and imperative for successful clinical outcomes. The current review discusses the current evidence and practical guidelines for prevention, early detection and management of TSE-related AEs.

Computer-Assisted Processing of Current Step Signals in Single Blocking Impact Electrochemistry.

Current step signals related to single-entity collisions in blocking impact electrochemistry were analyzed by computer-assisted processing for estimating the size distributions of various particles. In this work, three different types of entities were studied by single blocking impact electrochemistry: polystyrene nanospheres (350 nm diameter) and microspheres (1 μm diameter), phospholipid liposomes (300 nm diameter) and two different strains of Gram-negative bacillus bacteria (Escherichia coli and Shewanella oneidensis). The size estimations of these different entities from the current step signal analysis were compared and discussed according to the shape and size of each entity. From the magnitude of the current step transient, the size distribution of each entity was calculated by a new computer program assisting in the detection and analysis of single impact events in chronoamperometry measurements. The data processing showed that the size distributions obtained from the electrochemical data agreed with the dynamic light scattering and atomic force microscopy data for nanospheres and liposomes. In contrast, the size estimation calculated from the electrochemical data was underestimated for microspheres and bacteria. We demonstrated that our computer program was efficient for detecting and analyzing the collision events in single blocking impact electrochemistry for various entities from spherical hard nanoparticles to micrometer-sized rod-shaped living bacteria.

Forecasting High-Risk Behavioral and Medical Events in Children with Autism Using Digital Behavioral Records.

Individuals with Autism Spectrum Disorder may display interfering behaviors that limit their inclusion in educational and community settings, negatively impacting their quality of life. These behaviors may also signal potential medical conditions or indicate upcoming high-risk behaviors. This study explores behavior patterns that precede high-risk, challenging behaviors or seizures the following day. We analyzed an existing dataset of behavior and seizure data from 331 children with profound ASD over nine years. We developed a deep learning-based algorithm designed to predict the likelihood of aggression, elopement, and self-injurious behavior (SIB) as three high-risk behavioral events, as well as seizure episodes as a high-risk medical event occurring the next day. The proposed model attained accuracies of 78.4%, 80.68%, 85.43%, and 69.95% for predicting the next-day occurrence of aggression, SIB, elopement, and seizure episodes, respectively. The results were proven significant for more than 95% of the population for all high-risk event predictions using permutation-based statistical tests. Our findings emphasize the potential of leveraging historical behavior data for the early detection of high-risk behavioral and medical events, paving the way for behavioral interventions and improved support in both social and educational environments.

Power system event detection using Teager Kaiser Energy Operator and autoencoder based algorithm

As the global demand for electricity increases, modern power systems are operating closer to their stability limits. This increases power system events that vary in intensity. It is essential to detect these power system events as early as possible and to take the necessary control actions to prevent escalation. This paper proposes a power system event detection method based on the Teager Kaiser Energy Operator (TKEO) and Autoencoder. In the proposed method, the rules-based algorithm performs primary detection, and the detected events are confirmed using the TKEO and autoencoder-based algorithms. The effectiveness of the proposed method was tested using test signals, simulated power system signals, and real-time power system data obtained from an actual power system. The results reveal the robustness and accuracy of the proposed method in detecting single and multiple power system events in the PMU data.

The Value of Social Media Analysis for Adverse Events Detection and Pharmacovigilance: Scoping Review.

Adverse drug events pose an enormous public health burden, leading to hospitalization, disability, and death. Even the adverse events (AEs) categorized as nonserious can severely impact on patient's quality of life, adherence, and persistence. Monitoring medication safety is challenging. Web-based patient reports on social media may be a useful supplementary source of real-world data. Despite the growth of sophisticated techniques for identifying AEs using social media data, a consensus has not been reached as to the value of social media in relation to more traditional data sources. This study aims to evaluate and characterize the utility of social media analysis in adverse drug event detection and pharmacovigilance as compared with other data sources (such as spontaneous reporting systems and the clinical literature). In this scoping review, we searched 11 bibliographical databases and Google Scholar, followed by handsearching and forward and backward citation searching. Each record was screened by 2 independent reviewers at both the title and abstract stage and the full-text screening stage. Studies were included if they used any type of social media (such as Twitter or patient forums) to detect AEs associated with any drug medication and compared the results ascertained from social media to any other data source. Study information was collated using a piloted data extraction sheet. Data were extracted on the AEs and drugs searched for and included; the methods used (such as machine learning); social media data source; volume of data analyzed; limitations of the methodology; availability of data and code; comparison data source and comparison methods; results, including the volume of AEs, and how the AEs found compared with other data sources in their seriousness, frequencies, and expectedness or novelty (new vs known knowledge); and conclusions. Of the 6538 unique records screened, 73 publications representing 60 studies with a wide variety of extraction methods met our inclusion criteria. The most common social media platforms used were Twitter and online health forums. The most common comparator data source was spontaneous reporting systems, although other comparisons were also made, such as with scientific literature and product labels. Although similar patterns of AE reporting tended to be identified, the frequencies were lower in social media. Social media data were found to be useful in identifying new or unexpected AEs and in identifying AEs in a timelier manner. There is a large body of research comparing AEs from social media to other sources. Most studies advocate the use of social media as an adjunct to traditional data sources. Some studies also indicate the value of social media in understanding patient perspectives such as the impact of AEs, which could be better explored. RR2-10.2196/47068.

Comprehensive detection and dissection of interlineage recombination events in the SARS-CoV-2 pandemic.

The global prevalence of the XBB lineage presents a formidable challenge posed by the recombinant SARS-CoV-2 virus. The understanding of SARS-CoV-2's recombination preference assumes utmost significance in predicting future recombinant variants and adequately preparing for subsequent pandemics. Thus, an urgent need arises to establish a comprehensive landscape concerning SARS-CoV-2 recombinants worldwide and elucidate their evolutionary mechanisms. However, the initial step, involving the detection of potential recombinants from a vast pool of over 10 million sequences, presents a significant obstacle. In this study, we present CovRecomb, a lightweight methodology specifically designed to effectively identify and dissect interlineage SARS-CoV-2 recombinants. Leveraging CovRecomb, we successfully detected 135,567 putative recombinants across the entirety of 14.5 million accessed SARS-CoV-2 genomes. These putative recombinants could be classified into 1451 distinct recombination events, of which 206 demonstrated transmission spanning multiple countries, continents, or globally. Hotspot regions were identified in six specific areas, with prominence observed in the latter halves of the N-terminal domain and receptor-binding domain within the spike (S) gene. Epidemiological investigations revealed extensive recombination events occurring among different SARS-CoV-2 (sub)lineages, independent of lineage prevalence frequencies.

Fishing event detection and species classification using computer vision and artificial intelligence for electronic monitoring

Fisheries regulations require detailed catch reporting on commercial fishing vessels. Vital components for the sustainable management of fish stocks include a robust estimate of the number of fish caught and the species composition. Catch recording is often done manually by human observers on fishing vessels. Human observers are costly, and consistent data streams can be subject to observer availability and the weather. On-vessel cameras (electronic monitoring, EM) are a growing alternative to human observers. However, on-land human auditors are required to review hundreds of hours of videos recorded during fishing trips that can last for weeks. In this paper, a framework is presented to automatically detect fish in EM videos, count the total fishing events, and classify the fish species. For this purpose, a deep learning and computer vision-based model is developed to efficiently detect fish and fishers onboard a vessel. Secondly, a vision-based tracking pipeline tracks the detected fish and counts the total fishing events in the videos. Thirdly, the extracted fishing events are classified through a deep learning-based fish species classifier, to provide the distribution of different fish species caught for a fishing trip. For our experiments, the datasets were prepared using the electronic monitoring data of multiple fishing trips of a fishing vessel. The videos were recorded on Australian longline vessels targeting tunas and billfish. For the fish detection task, video frames were extracted and labelled manually to provide a digital ground-truth. For the fish species classification task, hundreds of fish images of multiple species were cropped to provide a training dataset for the fish classifier. For the fish counting task, manual counts for the fishing events of individual fish species were generated for the test fishing trips. The developed fish and fisher detector achieves a mean Average Precision of 87.0 % for fish and 94.0 % for fishers on test video frames. The fishing event detection pipeline achieves an Average Precision of 81.0 % and an Average Recall of 74.5 % on test videos. The fish species classifier achieves an Accuracy (Top-1) of 91.11 % for the classification of cropped fish images and 89.05 % for the classification of extracted fishing events from the videos. Experimental results show that our proposed computer vision and artificial intelligence-based solution for video analysis has great potential to automate the auditing process from electronic monitoring footage and contribute to the sustainable management of fish stocks.

Analysis of health related event detection in big data for physical education training movement detection.

Physical education and training are essential ways to improve the physical quality of the nation, and China has incorporated "building a healthy China" and "fitness for all" into its national development strategy, integrating a strong sports nation into the Chinese dream. The study of digital recording and automated training in sports is of profound value. Motion capture technology can digitally record the training process in a digital physical education training system. At the same time, accurate modeling and calculation can analyze the training effects and give appropriate guidance and feedback. This study develops a new and improved hierarchical K-means algorithm by combining the known classification algorithm K-means with a hierarchical algorithm. The performance of the old and new algorithms are compared and then applied to physical education training data to produce clustering results and analysis to reduce the model, which is used to reduce the number of parameters in the model and improve the recognition speed. The experimental results demonstrate that the relevant models proposed in this study achieve an average accuracy of 91.27% and 92.26%, respectively, which is better than a single network model and can effectively use big data for health event detection. The empirical results show that the improved model algorithm outperforms the single network model and can detect health events using big data.

The Role of Artificial Intelligence and Machine Learning in Cardiovascular Imaging and Diagnosis.

Cardiovascular diseases remain the leading cause of global mortality, underscoring the critical need for accurate and timely diagnosis. This narrative review examines the current applications and future potential of artificial intelligence (AI) and machine learning (ML) in cardiovascular imaging. We discuss the integration of these technologies across various imaging modalities, including echocardiography, computed tomography, magnetic resonance imaging, and nuclear imaging techniques. The review explores AI-assisted diagnosis in key areas such as coronary artery disease detection, valve disorders assessment, cardiomyopathy classification, arrhythmia detection, and prediction of cardiovascular events. AI demonstrates promise in improving diagnostic accuracy, efficiency, and personalized care. However, significant challenges persist, including data quality standardization, model interpretability, regulatory considerations, and clinical workflow integration. We also address the limitations of current AI applications and the ethical implications of their implementation in clinical practice. Future directions point towards advanced AI architectures, multimodal imaging integration, and applications in precision medicine and population health management. The review emphasizes the need for ongoing collaboration between clinicians, data scientists, and policymakers to realize the full potential of AI in cardiovascular imaging while ensuring ethical and equitable implementation. As the field continues to evolve, addressing these challenges will be crucial for the successful integration of AI technologies into cardiovascular care, potentially revolutionizing diagnostic capabilities and improving patient outcomes.

Role of quantitative imaging biomarkers in an early FDG-PET/CT for detection of immune-related adverse events in melanoma patients: a prospective study.

To evaluate the role of the novel quantitative imaging biomarker (QIB) SUVX% of 18F-FDG uptake extracted from early 18F-FDG-PET/CT scan at 4 weeks for the detection of immune-related adverse events (rAE) in a cohort of patients with metastatic melanoma (mM) patients receiving immune-checkpoint inhibitors (ICI). In this prospective non-interventional, one-centre clinical study, patients with mM, receiving ICI treatment, were regularly followed by 18F-FDG PET/CT. Patients were scanned at baseline, early point at week four (W4), week sixteen (W16) and week thirty-two (W32) after ICI initiation. A convolutional neural network (CNN) was used to segment three organs: lung, bowel, thyroid. QIB of irAE - SUVX% - was analyzed within the target organs and correlated with the clinical irAE status. Area under the receiver-operating characteristic curve (AUROC) was used to quantify irAE detection performance. A total of 242 18F-FDG PET/CT images of 71 mM patients were prospectively collected and analysed. The early W4 scan showed improved detection only for the thyroid gland compared to W32 scan (p=0.047). The AUROC for detection of irAE in the three target organs was highest when SUVX% was extracted from W16 scan and was 0.76 for lung, 0.53 for bowel and 0.81 for thyroid. SUVX% extracted from W4 scan did not improve detection of irAE compared to W16 scan (lung: p = 0.54, bowel: p = 0.75, thyroid: p = 0.3, DeLong test), as well as compared to W32 scan in lungs (p = 0.32) and bowel (p = 0.3). Early time point 18F-FDG PET/CT at W4 did not lead to statistically significant earlier detection of irAE. However, organ 18F-FDG uptake as quantified by SUVX% proved to be a consistent QIB of irAE. To better assess the role of 18F-FDG PET/CT in irAE detection, the time evolution of 18F-FDG PET/CT quantifiable inflammation would be of essence, only achievable in multi centric studies.

The spatiotemporal analysis of SARS-CoV-2 transmission in China since the termination of the dynamic zero-COVID policy

China's dynamic zero-COVID policy has effectively curbed the spread of SARS-CoV-2, while inadvertently creating immunity gaps within its population. Subsequent surges in COVID-19 cases linked to various SARS-CoV-2 lineages post-policy termination necessitate a thorough investigation into the epidemiological landscape. This study addresses this issue by analyzing a comprehensive dataset of 39,456 high-quality genomes collected nationwide over an 11-month period since policy termination. Through lineage assignment, phylogenetic analysis, pandemic pattern comparison, phylodynamic reconstruction, and recombination detection, we found that China's post-epidemic period could be divided into three stages, along with dynamic changes in dominant lineages. Geographical clustering of similar lineages implies the importance of cross-border cooperation among neighboring regions. Compared to the USA, UK, and Japan, China exhibits unique trajectories of lineage epidemics, characterized by initial lagging followed by subsequent advancement, indicating the potential influence of diverse prevention and control policies on lineage epidemic patterns. Hong Kong, Shanghai, and Hubei emerge as pivotal nodes in the nationwide spread, marking a shift in the transmission center from east to central regions of China. Although China hasn't experienced significant variant emergence, the detection and validation of the novel recombination event, XCN lineage, underscore the ongoing virus evolution. Overall, this study systematically analyzes the spatiotemporal transmission of SARS-CoV-2 virus in China since the termination of the dynamic zero-COVID policy, offering valuable insights for regional surveillance and evidence-based public health policymaking.

A high‐frequency greenhouse gas flux analysis tool: Insights from automated non‐steady‐state transparent soil chambers

AbstractNon‐steady‐state chambers are widely employed for quantifying soil emissions of CO2, CH4, and N2O. Automated non‐steady‐state (a‐NSS) soil chambers, when coupled with online gas analysers, offer the ability to capture high‐frequency measurements of greenhouse gas (GHG) fluxes. While these sampling systems provide valuable insights into GHG emissions, they present post‐measurement challenges, including the management of extensive datasets, intricate flux calculations, and considerations for temporal upscaling. In this study, a computationally efficient algorithm was developed to compute instantaneous fluxes and estimate diel flux patterns using continuous, high‐resolution data obtained from an a‐NSS sampling system. Applied to a 38‐day dataset, the algorithm captured concurrent field measurements of CO2, CH4, and N2O fluxes. The automated sampling system enables the acquisition of high‐frequency data, allowing the detection of episodic gas flux events. By using shape‐constrained additive models, a median percentage deviation (bias) of −1.031 and −4.340% was achieved for CO2 and N2O fluxes, respectively. Simpson's rule allowed for efficient upscale from instantaneous to diel flux values. As a result, the proposed algorithm can rapidly and simultaneously calculate CO2, CH4, and N2O fluxes, providing both instantaneous and diel values directly from raw, high‐temporal‐resolution data. These advancements significantly contribute to the field of GHG flux measurement, enhancing both the efficiency and accuracy of calculations for a‐NSS soil chambers and deepening our understanding of GHG emissions and their temporal dynamics.

Scalability and performance of decision tree for cardiovascular disease prediction

<span lang="EN-US">As one of the most common types of disease, cardiovascular disease is a serious health concern worldwide. Early detection is crucial for successful treatment and improved survival rates. The decision tree is a robust classifier for predicting the risk of cardiovascular disease and getting insights that would assist in making clinical decisions. However, selecting a better model for cardiovascular disease could be challenging due to scalability issues. Hence, this study examines the scalability and performance of decision trees for cardiovascular disease prediction. The study evaluated the performance of a decision tree for predicting cardiovascular disease. The performance evaluation was carried out by employing a confusion matrix, cross-validation score, model complexity, and training score for varying sizes of training samples. The experiment depicted that, the decision tree model was 88.8% accurate in predicting the presence or absence of cardiovascular disease. Therefore, the implementation of the decision tree is beneficial for the prediction and early detection of heart disease events in patients.</span>

Assessing two IMU-based gait event detection methods and their effect on spatiotemporal gait parameters across young and elderly populations

Assessing two IMU-based gait event detection methods and their effect on spatiotemporal gait parameters across young and elderly populations