Neural Features Research Articles

Prognostic Significance and Associations of Neural Network-Derived Electrocardiographic Features.

Subtle, prognostically important ECG features may not be apparent to physicians. In the course of supervised machine learning, thousands of ECG features are identified. These are not limited to conventional ECG parameters and morphology. We aimed to investigate whether neural network-derived ECG features could be used to predict future cardiovascular disease and mortality and have phenotypic and genotypic associations. We extracted 5120 neural network-derived ECG features from an artificial intelligence-enabled ECG model trained for 6 simple diagnoses and applied unsupervised machine learning to identify 3 phenogroups. Using the identified phenogroups, we externally validated our findings in 5 diverse cohorts from the United States, Brazil, and the United Kingdom. Data were collected between 2000 and 2023. In total, 1 808 584 patients were included in this study. In the derivation cohort, the 3 phenogroups had significantly different mortality profiles. After adjusting for known covariates, phenogroup B had a 20% increase in long-term mortality compared with phenogroup A (hazard ratio, 1.20 [95% CI, 1.17-1.23]; P<0.0001; phenogroup A mortality, 2.2%; phenogroup B mortality, 6.1%). In univariate analyses, we found phenogroup B had a significantly greater risk of mortality in all cohorts (log-rank P<0.01 in all 5 cohorts). Phenome-wide association study showed phenogroup B had a higher rate of future atrial fibrillation (odds ratio, 2.89; P<0.00001), ventricular tachycardia (odds ratio, 2.00; P<0.00001), ischemic heart disease (odds ratio, 1.44; P<0.00001), and cardiomyopathy (odds ratio, 2.04; P<0.00001). A single-trait genome-wide association study yielded 4 loci. SCN10A, SCN5A, and CAV1 have roles in cardiac conduction and arrhythmia. ARHGAP24 does not have a clear cardiac role and may be a novel target. Neural network-derived ECG features can be used to predict all-cause mortality and future cardiovascular diseases. We have identified biologically plausible and novel phenotypic and genotypic associations that describe mechanisms for the increased risk identified.

Synergistic Modeling of Liquid Properties: Integrating Neural Network-Derived Molecular Features with Modified Kernel Models.

A significant challenge in applying machine learning to computational chemistry, particularly considering the growing complexity of contemporary machine learning models, is the scarcity of available experimental data. To address this issue, we introduce an approach that derives molecular features from an intricate neural network-based model and applies them to a simpler conventional machine learning model that is robust to overfitting. This method can be applied to predict various properties of a liquid system, including viscosity or surface tension, based on molecular features drawn from the ab initio calculated free energy of solvation. Furthermore, we propose a modified kernel model that includes Arrhenius temperature dependence to incorporate theoretical principles and diminish extreme nonlinearity in the model. The modified kernel model demonstrated significant improvements in certain scenarios and possible extensions to various theoretical concepts of molecular systems.

Optimal Feature Selection and Classification for Parkinson’s Disease Using Deep Learning and Dynamic Bag of Features Optimization

Parkinson’s Disease (PD) is a neurological condition that worsens with time and is characterized bysymptoms such as cognitive impairment andbradykinesia, stiffness, and tremors. Parkinson’s is attributed to the interference of brain cells responsible for dopamine production, a substance regulating communication between brain cells. The brain cells involved in dopamine generation handle adaptation and control, and smooth movement. Convolutional Neural Networks are used to extract distinctive visual characteristics from numerous graphomotor sample representations generated by both PD and control participants. The proposed method presents an optimal feature selection technique based on Deep Learning (DL) and the Dynamic Bag of Features Optimization Technique (DBOFOT). Our method combines neural network-based feature extraction with a strong optimization technique to dynamically choose the most relevant characteristics from biological data. Advanced DL architectures are then used to classify the chosen features, guaranteeing excellent computational efficiency and accuracy. The framework’s adaptability to different datasets further highlights its versatility and potential for further medical applications. With a high accuracy of 0.93, the model accurately identifies 93% of the cases that are categorized as Parkinson’s. Additionally, it has a recall of 0.89, which means that 89% of real Parkinson’s patients are accurately identified. While the recall for Class 0 (Healthy) is 0.75, meaning that 75% of the real healthy cases are properly categorized, the precision decreases to 0.64 for this class, indicating a larger false positive rate.

AdaGIN: Adaptive Graph Interaction Network for Click-Through Rate Prediction

The goal of click-through rate (CTR) prediction in recommender systems is to effectively work with input features. However, existing CTR prediction models face three main issues. First, many models use a basic approach for feature combinations, leading to noise and reduced accuracy. Second, there is no consideration for the varying importance of features in different interaction orders, affecting model performance. Third, current model architectures struggle to capture different interaction signals from various semantic spaces, leading to sub-optimal performance. To address these issues, we propose the Adaptive Graph Interaction Network (AdaGIN) with the Graph Neural Networks-based Feature Interaction Module (GFIM), the Multi-semantic Feature Interaction Module (MFIM), and the Negative Feedback-based Search (NFS) algorithm. GFIM explicitly aggregates information between features and assesses their importance, while MFIM captures information from different semantic spaces. NFS uses negative feedback to optimize model complexity. Experimental results show AdaGIN outperforms existing models on large-scale public benchmark datasets.

Making the most of errors: Utilizing erroneous classifications generated by machine-learning models of neuroimaging data to capture disorder heterogeneity.

Within-disorder heterogeneity complicates mapping the neurobiological features of psychopathology to Diagnostic and Statistical Manual of Mental Disorders conceptualizations. The present study explored the patterns of diagnostic classification errors among disorders with commonly co-occurring features to examine this heterogeneity. Classification analyses were conducted with the University of California, Los Angeles Phenomics Study database using a support-vector classifier to differentiate disorders via whole brain task-based functional connectivity, predicting that model misclassifications would provide insight about brain connectivity characteristics shared across disorders. Whether symptoms and specific brain networks could account for misclassification rates was also explored. The classification model performed better than chance (44% accuracy, p = .01) and revealed that misclassification of schizophrenia (SCZ) as bipolar disorder (BD; 38%) and BD as SCZ (36%) was symmetrical. Attention-deficit/hyperactivity disorder (ADHD) was misclassified as BD at the highest rate (46%) and higher than the converse (17%). SCZ and ADHD were misclassified least (15% SCZ as ADHD and 22% ADHD as SCZ). Considerable variance in misclassification of SCZ as BD (R2 = .83) and BD as SCZ (R2 = .71) could be accounted for by symptoms of both SCZ and BD. Permutation testing revealed disorder- and network-specific effects, with certain networks improving classification accuracy and others hindering it for specific disorders. An approach focused on classification errors replicated known disorder overlap, producing errors in the expected configuration. Further, it identified clinical and neural features within and across diagnostic categories that contribute to disorder misclassification and within-disorder heterogeneity. This approach may facilitate neurobiologically informed phenotypic differentiation within diagnostic groups. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Integrating temporal deep learning models for predicting screen-out risk levels in hydraulic fracturing

Amid the transformative shift in global energy structures, the exploitation and utilization of shale gas, an essential unconventional natural gas resource, have drawn widespread attention from both industrial and academic circles. However, screen-out incidents during hydraulic fracturing operations pose significant obstacles to extraction efficiency and safety. Traditional prediction methods, which rely on empirical estimations and simplified models, are deficient in accuracy and real-time applicability. Addressing this, our study introduces a novel deep learning ensemble integrating Gated Recurrent Units (GRU), Transformer, and One-Dimensional Convolutional Neural Networks (1D-CNN) for precise screen-out prediction. This approach markedly improves predictive accuracy by efficiently processing time-series data and capturing the complex dynamics of fracturing processes. Furthermore, the application of the correlation coefficient method and random forest algorithm for feature selection optimizes model input and further enhances prediction accuracy and operational efficiency. Our comparative analysis demonstrates the model’s superiority, achieving an F1 score of 0.951 and a loss of 0.430, clearly surpassing traditional and other deep learning methods. This integration of advanced neural architectures and feature selection techniques not only advances screen-out prediction but also yields practical insights for optimizing shale gas extraction strategies and enhancing safety.

Characterizing Subsurface Structures From Hard and Soft Data With Multiple‐Condition Fusion Neural Network

AbstractAccurately inferring realistic subsurface structures poses a considerable challenge due to the impact of morphology on flow and transport behaviors. Traditional subsurface characterization relies on two primary types of data: hard data, derived from direct subsurface measurements, and soft data, encompassing remotely sensed geophysical information and its interpretation. Existing deep‐learning‐based methodologies predominantly focus on the transition from multiple observations to subsurface structures. However, implicit non‐linear correlations among diverse data sources often remain underutilized, leading to potential bias and errors. In this study, we introduce a multiple‐condition fusion network (MCF‐Net) to characterize subsurface structures based on both hard and soft data. To harness the full potential of multiple‐source subsurface observations, two distinct neural networks extract implicit features from hard and soft data. The integration of these features is achieved through multiple‐condition fusion blocks, designed to capture representative characteristics. These blocks are also adept at reconstructing heterogeneous structures and facilitating hydrological parameterization. MCF‐Net exhibits accuracy in estimating subsurface structures across various types of subsurface observations. Experimental results underscore the utility and superiority of MCF‐Net in applications of hydrogeological modeling.

Reference-based In-loop Filter with Robust Neural Feature Transfer for Video Coding

Reference-based In-loop Filter with Robust Neural Feature Transfer for Video Coding

Deep feature response discriminative calibration

Deep neural networks (DNNs) have numerous applications across various domains. Several optimization techniques, such as ResNet and SENet, have been proposed to improve model accuracy. These techniques improve the model performance by adjusting or calibrating feature responses according to a uniform standard. However, they lack the discriminative calibration for different features, thereby introducing limitations in the model output. Therefore, we propose a method that discriminatively calibrates feature responses. The preliminary experimental results indicate that the neural feature response follows a Gaussian distribution. Consequently, we compute confidence values by employing the Gaussian probability density function, and then integrate these values with the original response values. The objective of this integration is to improve the feature discriminability of the neural feature response. Based on the calibration values, we propose a plugin-based calibration module incorporated into a modified ResNet architecture, termed Response Calibration Networks (ResCNet). Extensive experiments on datasets like CIFAR-10, CIFAR-100, SVHN, and ImageNet demonstrate the effectiveness of the proposed approach.

Deep Learning in Spinal Endoscopy: U-Net Models for Neural Tissue Detection

Biportal endoscopic spine surgery (BESS) is minimally invasive and therefore benefits both surgeons and patients. However, concerning complications include dural tears and neural tissue injuries. In this study, we aimed to develop a deep learning model for neural tissue segmentation to enhance the safety and efficacy of endoscopic spinal surgery. We used frames extracted from videos of 28 endoscopic spine surgeries, comprising 2307 images for training and 635 images for validation. A U-Net-like architecture is employed for neural tissue segmentation. Quantitative assessments include the Dice-Sorensen coefficient, Jaccard index, precision, recall, average precision, and image-processing time. Our findings revealed that the best-performing model achieved a Dice-Sorensen coefficient of 0.824 and a Jaccard index of 0.701. The precision and recall values were 0.810 and 0.839, respectively, with an average precision of 0.890. The model processed images at 43 ms per frame, equating to 23.3 frames per second. Qualitative evaluations indicated the effective identification of neural tissue features. Our U-Net-based model robustly performed neural tissue segmentation, indicating its potential to support spine surgeons, especially those with less experience, and improve surgical outcomes in endoscopic procedures. Therefore, further advancements may enhance the clinical applicability of this technique.

Prediction of the Remaining Useful Life of Bearings Through CNN-Bi-LSTM-Based Domain Adaptation Model.

Predicting the remaining useful life (RUL) of mechanical bearings is crucial in the industry. Estimating the RUL enables the assessment of health bearing, maintenance planning, and significant cost reduction, thereby fostering industrial development. Existing models rely on traditional feature engineering with feature changes because operating conditions pose a major challenge to the generalization of RUL prediction models. This study focuses on neural network-based feature engineering and the downstream prediction of the RUL, eliminating the need for specific prior knowledge and simplifying the development and maintenance of models. Initially, a convolutional neural network (CNN) model is employed for feature engineering. Subsequently, a bidirectional long short-term memory network (Bi-LSTM) model is used to capture the time-series degradation characteristics of the engineered features and predict the RUL through regression. Finally, the study examines the influence of operating conditions in the model and integrates domain adaptation to minimize differences in feature distribution, thereby enhancing the model's generalizability for the RUL prediction.

Development of an Artificial Intelligence System for Distinguishing Malignant from Benign Soft Tissue Tumors Using Contrast-Enhanced MR Images.

The integration of artificial intelligence (AI) into orthopedics has enhanced the diagnosis of various conditions; however, its use in diagnosing soft-tissue tumors remains limited owing to its complexity. This study aimed to develop and assess an AI-driven diagnostic support system for magnetic resonance imaging (MRI)-based soft-tissue tumor diagnosis, potentially improving accuracy and aiding radiologists and orthopedic surgeons. Experienced orthopedic oncologists and radiologists annotated 720 images from 77 cases (41 benign and 36 malignant soft-tissue tumors). Eleven tumor subtypes were identified and classified into benign and malignant groups based on histological diagnosis. Utilizing the standard machine learning classifier pipeline, we examined and down-selected imaging protocols and their predominant radiomic features within the tumor's three-dimensional region to differentiate between benign and malignant tumors. Among the scan protocols, contrast-enhanced T1 weighted fat-suppressed images showed the most accurate classification based on radiomics features. We focused on the two-dimensional features from the largest tumor boundary surface and its neighboring slices, leveraging texture-based radiomic and deep convolutional neural network features from a pretrained VGG19 model. The test data comprised 44 contrast-enhanced images (22 benign and 22 malignant soft-tissue tumors) containing six malignant and five benign subtypes distinct from the training data. We compared expert and non-expert human performances against AI by assessing malignancy detection and the time required for classification. The AI model showed comparable accuracy (AUC 0.91) to that of radiologists (AUC 0.83) and orthopedic surgeons (AUC 0.73). Notably, the AI model processed data approximately 400 times faster than its human counterparts, showcasing its capacity to significantly boost diagnostic efficiency. We developed an AI-driven diagnostic support system for MRI-based soft-tissue tumor diagnosis. While additional refinement is necessary for clinical applications, our system has exhibited promising potential in differentiating between benign and malignant soft-tissue tumors based on MRI.

Forecasting future earthquakes with deep neural networks: application to California

SUMMARY We use the spatial map of the logarithm of past estimated released earthquake energies as input of fully convolutional networks (FCN) to forecast future earthquakes. This model is applied to California and compared with an elaborated version of the epidemic type aftershock sequence (ETAS) model. Our long-term earthquake forecast simulations show that the FCN model is close to the ETAS model in forecasting earthquakes with $M \ge 3.0,\,\,4.0,\,\,{\rm{and\,\,}}5.0$ according to the Molchan diagram. Moreover, training and implementing the FCN model is 2000–4000 times faster than calibrating the ETAS model and generating its probabilistic forecasts. The FCN model is straightforward in terms of its neural network structure and feature engineering. It does not require extensive knowledge of statistical seismology or the analysis of earthquake catalogue completeness. Using the earthquake catalogue with $M \ge 0$ as FCN input can enhance the model's performance in some time–magnitude forecasting windows.

Image Stitching of Low-Resolution Retinography Using Fundus Blur Filter and Homography Convolutional Neural Network

Great advances in stitching high-quality retinal images have been made in recent years. On the other hand, very few studies have been carried out on low-resolution retinal imaging. This work investigates the challenges of low-resolution retinal images obtained by the D-EYE smartphone-based fundus camera. The proposed method uses homography estimation to register and stitch low-quality retinal images into a cohesive mosaic. First, a Siamese neural network extracts features from a pair of images, after which the correlation of their feature maps is computed. This correlation map is fed through four independent CNNs to estimate the homography parameters, each specializing in different corner coordinates. Our model was trained on a synthetic dataset generated from the Microsoft Common Objects in Context (MSCOCO) dataset; this work added an important data augmentation phase to improve the quality of the model. Then, the same is evaluated on the FIRE retina and D-EYE datasets for performance measurement using the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM). The obtained results are promising: the average PSNR was 26.14 dB, with an SSIM of 0.96 on the D-EYE dataset. Compared to the method that uses a single neural network for homography calculations, our approach improves the PSNR by 7.96 dB and achieves a 7.86% higher SSIM score.

Schizophrenia pathology reverse-translated into mouse shows hippocampal hyperactivity, psychosis behaviors and hyper-synchronous events.

Decades of research into the function of the medial temporal lobe has driven curiosity around clinical outcomes associated with hippocampal dysfunction, including psychosis. Post-mortem analyses of brain tissue from human schizophrenia brain show decreased expression of the NMDAR subunit GluN1 confined to the dentate gyrus with evidence of downstream hippocampal hyperactivity in CA3 and CA1. Little is known about the mechanisms of the emergence of hippocampal hyperactivity as a putative psychosis biomarker. We have developed a reverse-translation mouse to study critical neural features. We had previously studied a dentate gyrus (DG)-specific GluN1 KO, which displays hippocampal hyperactivity and a psychosis-relevant behavioral phenotype. Here, we expressed an inhibitory DREADD (pAAV-CaMKIIa-hM4D(Gi)-mCherry) in granule cells of the mouse dentate gyrus, and continuously inhibited the region for 21 days in adolescent (6 weeks) and adult (10 weeks) C57BL/6 J mice with DREADD agonist Compound 21 (C21). Following this period, we quantified activity in the hippocampal subfields by assessing cFos expression, hippocampally mediated behaviors, and hippocampal local field potential with an intracerebral probe with continual monitoring over time. DG inhibition during adolescence generates an increase in hippocampal activity in CA3 and CA1, impairs social cognition and spatial working memory, as well as shows evidence of increased activity in local field potentials as spontaneous synchronous bursts of activity, which we term hyper-synchronous events (HSEs) in hippocampus. The same DG inhibition delivered during adulthood in the mouse lacks these outcomes. These results suggest a sensitive period in development in which the hippocampus is susceptible to DG inhibition resulting in hippocampal hyperactivity and psychosis-like behavioral outcomes.

From Text to Safety: A Novel Framework for Mining Unsafe Aviation Events Using Advanced Neural Network and Feature Network

The rapid growth of the aviation industry highlights the need for strong safety management. Analyzing data on unsafe aviation events is crucial for preventing risks. This paper presents a new method that integrates the Transformer network model, clustering analysis, and feature network modeling to analyze Chinese text data on unsafe aviation events. Initially, the Transformer model is used to generate summaries of event texts, and the performance of three pre-trained Chinese models is evaluated and compared. Next, the Jieba tool is applied to segment both summarized and original texts to extract key features of unsafe events and prove the effectiveness of the pre-trained Transformer model in simplifying lengthy and redundant original texts. Then, cluster analysis based on text similarity categorizes the extracted features. By solving the correlation matrix of these features, this paper constructs a feature network for unsafe aviation events. The network’s global and individual metrics are calculated and then used to identify key feature nodes, which alert aviation professionals to focus more on the decision-making process for safety management. Based on the established network and these metrics, a data-driven hidden danger warning strategy is proposed and illustrated. Overall, the proposed method can effectively analyze Chinese texts of unsafe aviation events and provide a basis for improving aviation safety management.

Ensemble graph neural networks for fake news detection using user engagement and text features

In this digital world with massive information sharing, spreading misinformation is relatively easy, but its consequences are really enormous. Detecting fake news is a critical challenge in today's digital age, where misinformation can spread rapidly and influence public opinion. This paper proposes an ensemble model (EGNN) that leverages Graph Neural Networks (GNNs) and text features to enhance the accuracy of fake news detection. The model is designed to operate effectively even when labeled data is scarce by utilizing a substantial amount of unlabeled data. We employ multiple GNNs, including a standard GNN, a Graph Attention Network (GAT), and a Bidirectional GCN (BiGCN), to capture user engagement patterns and social context. Additionally, text embeddings are extracted using spaCy, BERT, and a combination of spaCy with user profile information. These embeddings are then classified using a neural network, and the final detection is performed using ensemble techniques such as Majority Voting, Weighted Average, and Stacking. To address class imbalance, we incorporate focal loss alongside the traditional binary cross-entropy loss. Experimental results demonstrate that our proposed EGNN model significantly improves the detection of fake news, providing a robust solution for this pervasive problem.

Modeling Upper Limb Rehabilitation-Induced Recovery after Stroke: The Role of Attention as a Clinical Confounder.

People who have survived stroke may have motor and cognitive impairments. High dose of motor rehabilitation was found to provide clinically relevant improvement to upper limb (UL) motor function. Besides, mounting evidence suggests that clinical, neural, and neurophysiological features are associated with spontaneous recovery. However, the association between these features and rehabilitation-induced, rather than spontaneous, recovery has never been fully investigated.The objective was to explore the association between rehabilitation dose and UL motor outcome after stroke, as well as to identify which variables can be considered potential candidate predictors of motor recovery. People who survived stroke were assessed before and after a period of rehabilitation using motor, cognitive, neuroanatomical, and neurophysiological measures. We investigated the association between dose of rehabilitation and UL response (ie, Fugl-Meyer Assessment for upper extremity [FMA-UE]), using ordinary least squares regression as the primary analysis. To obtain unbiased estimates, adjusting covariates were selected using a directed acyclic graph. Baseline FMA-UE was the only factor associated with motor recovery (b = 0.99; 95% CI = 0.83 to 1.15 points). Attention emerged as a confounder of the association between rehabilitation and final FMA-UE (b = 5.5; 95% CI = -0.8 to 11.9 points), influencing both rehabilitation and UL response. Preserved attention in people who have survived stroke might lead to greater UL motor recovery, albeit estimates have high levels of variability. Moreover, the increase in the dose of rehabilitation can lead to 5.5 points improvement on the FMA-UE, a nonsignificant but potentially meaningful finding. The approach described here discloses a new framework for investigating the effect of rehabilitation treatment as a potential driver of recovery. Attentional resources could play a key role in UL motor recovery. There is a potential association between amount of UL recovery and dose of rehabilitation delivered, needing further exploration. Preserved attention and rehabilitation dose are candidate predictors of UL motor recovery.

Spark in the darkness: Discovering action potentials in brain tumors

Gliomas exhibit significant molecular diversity and poor prognosis. In this issue of Cancer Cell, Curry etal. apply Patch-seq on human glioma samples uncovering hybrid cells with glial and neuronal features, capable of firing action potentials in isocitrate dehydrogenase mutant gliomas. These findings highlight the importance of neural features in tumor biology and progression.

Methods for Corrosion Detection in Pipes Using Thermography: A Case Study on Synthetic Datasets

This study reviews advanced methods for corrosion detection and characterization in pipes using thermography, with a focus on addressing the limitations posed by small datasets. Thermography captures temperature distributions on the surface of pipes to identify subsurface defects. The challenges of sequential data processing, neural network performance, feature extraction, and dataset size are discussed, with proposed solutions such as advanced algorithms, feature selection techniques, and data augmentation. Given the significant gap in the current literature, there is a need for larger, more diverse datasets to train more robust and accurate machine learning models. A case study combining experimental data with Finite Element Method (FEM) simulations demonstrates that augmenting datasets with synthetic data significantly improves defect detection accuracy. These findings highlight the potential of integrating thermography with machine learning to enhance defect detection, providing insights for future research and practical applications.