RESUMEN
Graph neural networks (GNNs) have achieved state-of-the-art performance in graph representation learning. Message passing neural networks, which learn representations through recursively aggregating information from each node and its neighbors, are among the most commonly-used GNNs. However, a wealth of structural information of individual nodes and full graphs is often ignored in such process, which restricts the expressive power of GNNs. Various graph data augmentation methods that enable the message passing with richer structure knowledge have been introduced as one main way to tackle this issue, but they are often focused on individual structure features and difficult to scale up with more structure features. In this work we propose a novel approach, namely collective structure knowledge-augmented graph neural network (CoS-GNN), in which a new message passing method is introduced to allow GNNs to harness a diverse set of node- and graph-level structure features, together with original node features/attributes, in augmented graphs. In doing so, our approach largely improves the structural knowledge modeling of GNNs in both node and graph levels, resulting in substantially improved graph representations. This is justified by extensive empirical results where CoS-GNN outperforms state-of-the-art models in various graph-level learning tasks, including graph classification, anomaly detection, and out-of-distribution generalization.
RESUMEN
Complex-valued convolutional neural networks (CVCNNs) have been demonstrated effectiveness in classifying complex signals and synthetic aperture radar (SAR) images. However, due to the introduction of complex-valued parameters, CVCNNs tend to become redundant with heavy floating-point operations. Model sparsity is emerged as an efficient method of removing the redundancy without much loss of performance. Currently, there are few studies on the sparsity problem of CVCNNs. Therefore, a complex-valued soft-log threshold reweighting (CV-SLTR) algorithm is proposed for the design of sparse CVCNN to reduce the number of weight parameters and simplify the structure of CVCNN. On one hand, considering the difference between complex and real numbers, we redefine and derive the complex-valued log-sum threshold method. On the other hand, by considering the distinctive characteristics of complex-valued convolutional (CConv) layers and complex-valued fully connected (CFC) layers of CVCNNs, the complex-valued soft and log-sum threshold methods are respectively developed to prune the weights of different layers during the forward propagation, and the sparsity thresholds are optimized during the backward propagation by inducing a sparsity budget. Furthermore, different optimizers can be integrated with CV-SLTR. When stochastic gradient descent (SGD) is used, the convergence of CV-SLTR is proved if Lipschitzian continuity is satisfied. Experiments on the RadioML 2016.10A and S1SLC-CVDL datasets show that the proposed algorithm is efficient for the sparsity of CVCNNs. It is worth noting that the proposed algorithm has fast sparsity speed while maintaining high classification accuracy. These demonstrate the feasibility and potential of the CV-SLTR algorithm.
RESUMEN
BACKGROUND: To support dentists with limited experience, this study trained and compared six convolutional neural networks to detect crossbites and classify non-crossbite, frontal, and lateral crossbites using 2D intraoral photographs. METHODS: Based on 676 photographs from 311 orthodontic patients, six convolutional neural network models were trained and compared to classify (1) non-crossbite vs. crossbite and (2) non-crossbite vs. lateral crossbite vs. frontal crossbite. The trained models comprised DenseNet, EfficientNet, MobileNet, ResNet18, ResNet50, and Xception. FINDINGS: Among the models, Xception showed the highest accuracy (98.57%) in the test dataset for classifying non-crossbite vs. crossbite images. When additionally distinguishing between lateral and frontal crossbites, average accuracy decreased with the DenseNet architecture achieving the highest accuracy among the models with 91.43% in the test dataset. CONCLUSIONS: Convolutional neural networks show high potential in processing clinical photographs and detecting crossbites. This study provides initial insights into how deep learning models can be used for orthodontic diagnosis of malocclusions based on intraoral 2D photographs.
Asunto(s)
Aprendizaje Profundo , Maloclusión , Redes Neurales de la Computación , Humanos , Maloclusión/diagnóstico por imagen , Maloclusión/diagnóstico , Femenino , Masculino , Fotografía Dental/métodos , Fotograbar/métodos , AdolescenteRESUMEN
Artificial intelligence (AI) has had a significant impact on human life because of its pervasiveness across industries and its rapid development. Although AI has achieved superior performance in learning and reasoning, it encounters challenges such as substantial computational demands, privacy concerns, communication delays, and high energy consumption associated with cloud-based models. These limitations have facilitated a paradigm change in on-device AI processing, which offers enhanced privacy, reduced latency, and improved power efficiency through the direct execution of computations on devices. With advancements in neuromorphic systems, spiking neural networks (SNNs), often referred to as the next generation of AI, are currently in focus as on-device AI. These technologies aim to mimic the human brain efficiency and provide promising real-time processing with minimal energy. This study reviewed the application of SNNs in the analysis of biomedical signals (electroencephalograms, electrocardiograms, and electromyograms), and consequently, investigated the distinctive attributes and prospective future paths of SNNs models in the field of biomedical signal analysis.
RESUMEN
The integration of Spiking Neural Networks (SNNs) into the analysis and interpretation of physiological and speech signals has emerged as a groundbreaking approach, offering enhanced performance and deeper insights into the underlying biological processes. This review aims to summarize key advances, methodologies, and applications of SNNs within these domains, highlighting their unique ability to mimic the temporal dynamics and efficiency of the human brain. We dive into the core principles of SNNs, their neurobiological underpinnings, and the computational advantages they bring to signal processing, particularly in handling the temporal and spatial complexities inherent in physiological and speech data. Comparative analyses with conventional neural network models are presented to underscore the superior efficiency, lower power consumption, and higher temporal resolution of SNNs. The review further explores challenges and future prospects, highlighting the potential of SNNs to revolutionize wearable healthcare monitoring systems, neuroprosthetic devices, and natural language processing technologies. By providing a comprehensive overview of current strategies, this review aims to inspire innovative approaches in the field, fostering advances in real-time and energy-efficient processing of complex biological signals.
RESUMEN
Wildfires pose significant threats worldwide, requiring accurate prediction for mitigation. This study uses machine learning techniques to forecast wildfire severity in the Upper Colorado River basin. Datasets from 1984 to 2019 and key indicators like weather conditions and land use were employed. Random Forest outperformed Artificial Neural Network, achieving 72â¯% accuracy. Influential predictors include air temperature, vapor pressure deficit, NDVI, and fuel moisture. Solar radiation, SPEI, precipitation, and evapotranspiration also contribute significantly. Validation against actual severities from 2016 to 2019 showed mean prediction errors of 11.2â¯%, affirming the model's reliability. These results highlight the efficacy of machine learning in understanding wildfire severity, especially in vulnerable regions.
RESUMEN
Predicting protein-ligand binding affinity is a crucial and challenging task in structure-based drug discovery. With the accumulation of complex structures and binding affinity data, various machine-learning scoring functions, particularly those based on deep learning, have been developed for this task, exhibiting superiority over their traditional counterparts. A fusion model sequentially connecting a graph neural network (GNN) and a convolutional neural network (CNN) to predict protein-ligand binding affinity is proposed in this work. In this model, the intermediate outputs of the GNN layers, as supplementary descriptors of atomic chemical environments at different levels, are concatenated with the input features of CNN. The model demonstrates a noticeable improvement in performance on CASF-2016 benchmark compared to its constituent CNN models. The generalization ability of the model is evaluated by setting a series of thresholds for ligand extended-connectivity fingerprint similarity or protein sequence similarity between the training and test sets. Masking experiment reveals that model can capture key interaction regions. Furthermore, the fusion model is applied to a virtual screening task for a novel target, PI5P4Kα. The fusion strategy significantly improves the ability of the constituent CNN model to identify active compounds. This work offers a novel approach to enhancing the accuracy of deep learning models in predicting binding affinity through fusion strategies.
RESUMEN
PURPOSE: To obtain high-resolution velocity fields of cerebrospinal fluid (CSF) and cerebral blood flow by applying a physics-guided neural network (div-mDCSRN-Flow) to 4D flow MRI. METHODS: The div-mDCSRN-Flow network was developed to improve spatial resolution and denoise 4D flow MRI. The network was trained with patches of paired high-resolution and low-resolution synthetic 4D flow MRI data derived from computational fluid dynamic simulations of CSF flow within the cerebral ventricles of five healthy cases and five Alzheimer's disease cases. The loss function combined mean squared error with a binary cross-entropy term for segmentation and a divergence-based regularization term for the conservation of mass. Performance was assessed using synthetic 4D flow MRI in one healthy and one Alzheimer' disease cases, an in vitro study of healthy cerebral ventricles, and in vivo 4D flow imaging of CSF as well as flow in arterial and venous blood vessels. Comparison was performed to trilinear interpolation, divergence-free radial basis functions, divergence-free wavelets, 4DFlowNet, and our network without divergence constraints. RESULTS: The proposed network div-mDCSRN-Flow outperformed other methods in reconstructing high-resolution velocity fields from synthetic 4D flow MRI in healthy and AD cases. The div-mDCSRN-Flow network reduced error by 22.5% relative to linear interpolation for in vitro core voxels and by 49.5% in edge voxels. CONCLUSION: The results demonstrate generalizability of our 4D flow MRI super-resolution and denoising approach due to network training using flow patches and physics-based constraints. The mDCSRN-Flow network can facilitate MRI studies involving CSF flow measurements in cerebral ventricles and association of MRI-based flow metrics with cerebrovascular health.
RESUMEN
Globally, we are in the midst of a biodiversity crisis and megadiverse countries become key targets for conservation. South Africa, the only country in the world hosting three biodiversity hotspots within its borders, harbours a tremendous diversity of at-risk species deserving to be protected. However, the lengthy risk assessment process and the lack of required data to complete assessments is a serious limitation to conservation since several species may slide into extinction while awaiting risk assessment. Here, we employed a deep neural network model integrating species climatic and geographic features to predict the conservation status of 116 unassessed plant species. Our analysis involved in total of 1072 plant species and 96,938 occurrence points. The best-performing model exhibits high accuracy, reaching up to 83.6% at the binary classification and 56.8% at the detailed classification. Our best-performing model at the binary classification predicts that 32% (25 species) and 8% (3 species) of Data Deficient and Not-Evaluated species respectively, are likely threatened, amounting to a proportion of 24.1% of unassessed species facing a risk of extinction. Interestingly, all unassessed species predicted to be threatened are in protected areas, revealing the effectiveness of South Africa's network of protected areas in conservation, although these likely threatened species are more abundant outside protected areas. Considering the limitation in assessing only species with available data, there remains a possibility of a higher proportion of unassessed species being imperilled.
RESUMEN
Activated sludge models are increasingly being adopted to guide the operation of wastewater treatment plants. Chemical oxygen demand (COD) is an indispensable input for such models. To ensure that the activated sludge mathematical model can adapt to various water quality conditions and minimize prediction errors, it is essential to predict the parameters of the COD components in real-time based on the actual influent COD concentrations. However, conventional methods of determining the components' contributions are too intricate and time-consuming to be really useful. In this study, the chemical oxygen demand in the actual waste water treatment plant was disassembled and analyzed. The research involved determining the proportions of each COD component, assessing the reliability of the measurement parameters, and examining potential factors affecting measurement accuracy, including weather conditions, pipeline conditions, and residents' habits. Then, a backpropagation neural network was developed which can deliver real-time predictions for five important contributors to COD in real time. In addition, using the receiver operating characteristics curve and prediction accuracy to evaluate the performance of the prediction model. For all five components, which SS, XS, SI, XA, and XH, the prediction accuracy of model was more than 80 %. The maximum deviation values of these parameters fall within the range of the actual detected values, suggesting that the model's predictions align well with real-world observations, and demonstrated prediction performance adequate for practical application in wastewater treatment. This article can provide research basis for the engineering application of activated sludge model and help for the intelligent upgrading of waste water treatment plants.
RESUMEN
With the development of automated malware toolkits, cybersecurity faces evolving threats. Although visualization-based malware analysis has proven to be an effective method, existing approaches struggle with challenging malware samples due to alterations in the texture features of binary images during the visualization preprocessing stage, resulting in poor performance. Furthermore, to enhance classification accuracy, existing methods sacrifice prediction time by designing deeper neural network architectures. This paper proposes PAFE, a lightweight and visualization-based rapid malware classification method. It addresses the issue of texture feature variations in preprocessing through pixel-filling techniques and applies data augmentation to overcome the challenges of class imbalance in small sample datasets. PAFE combines multi-scale feature fusion and a channel attention mechanism, enhancing feature expression through modular design. Extensive experimental results demonstrate that PAFE outperforms the current state-of-the-art methods in both efficiency and effectiveness for malware variant classification, achieving an accuracy rate of 99.25 % with a prediction time of 10.04 ms.
RESUMEN
Over the last several years, the COVID-19 epidemic has spread over the globe. People have become used to the novel standard, which involves working from home, chatting online, and keeping oneself clean, to stop the spread of COVID-19. Due to this, many public spaces make an effort to make sure that their visitors wear proper face masks and maintain a safe distance from one another. It is impossible for monitoring workers to ensure that everyone is wearing a face mask; automated solutions are a far better option for face mask identification and monitoring to assist control public conduct and reduce the COVID-19 epidemic. The motivation for developing this technology was the need to identify those individuals who uncover their faces. Most of the previously published research publications focused on various methodologies. This study built new methods namely K-medoids, K-means, and Fuzzy K-Means(FKM) to use image pre-processing to get the better quality of the face and reduce the noise data. In addition, this study investigates various machine learning models Convolutional neural networks (CNN) with pre-trained (DenseNet201, VGG-16, and VGG-19) models, and Support Vector Machine (SVM) for the detection of face masks. The experimental results of the proposed method K-medoids with pre-trained model DenseNet201 achieved the 97.7 % accuracy best results for face mask identification. Our research results indicate that the segmentation of images may improve the identification of accuracy. More importantly, the face mask identification tool is more beneficial when it can identify the face mask in a side-on approach.
RESUMEN
BACKGROUND: Artificial intelligence (AI) models are being increasingly studied for the detection of variations and pathologies in different imaging modalities. Nasal septal deviation (NSD) is an important anatomical structure with clinical implications. However, AI-based radiographic detection of NSD has not yet been studied. OBJECTIVE: This research aimed to develop and evaluate a real-time model that can detect probable NSD using cone beam computed tomography (CBCT) images. METHODS: Coronal section images were obtained from 204 full-volume CBCT scans. The scans were classified as normal and deviated by 2 maxillofacial radiologists. The images were then used to train and test the AI model. Mask region-based convolutional neural networks (Mask R-CNNs) comprising 3 different backbones-ResNet50, ResNet101, and MobileNet-were used to detect deviated nasal septum in 204 CBCT images. To further improve the detection, an image preprocessing technique (contrast enhancement [CEH]) was added. RESULTS: The best-performing model-CEH-ResNet101-achieved a mean average precision of 0.911, with an area under the curve of 0.921. CONCLUSIONS: The performance of the model shows that the model is capable of detecting nasal septal deviation. Future research in this field should focus on additional preprocessing of images and detection of NSD based on multiple planes using 3D images.
Asunto(s)
Tomografía Computarizada de Haz Cónico , Tabique Nasal , Redes Neurales de la Computación , Prueba de Estudio Conceptual , Humanos , Tomografía Computarizada de Haz Cónico/métodos , Tabique Nasal/diagnóstico por imagen , Femenino , Masculino , Adulto , Persona de Mediana EdadRESUMEN
BACKGROUND AND AIM: Dental caries is largely preventable, yet an important global health issue. Numerous systematic reviews have summarized the efficacy of artificial intelligence (AI) models for the diagnosis and detection of dental caries. Therefore, this umbrella review aimed to synthesize the results of systematic reviews on the application and effectiveness of AI models in diagnosing and detecting dental caries. METHODS: MEDLINE/PubMed, IEEE Explore, Embase, and Cochrane Database of Systematic Reviews were searched to retrieve studies. Two authors independently screened the articles based on eligibility criteria and then, appraised the included articles. The findings are summarized in tabulation form and discussed using the narrative method. RESULT: A total of 1249 entries were identified out of which 7 were finally included. The most often employed AI algorithms were the multilayer perceptron, support vector machine (SVM), and neural networks. The algorithms were built to perform the segmentation, classification, caries detection, diagnosis, and caries prediction from several sources, including periapical radiographs, panoramic radiographs, smartphone images, bitewing radiographs, near-infrared light transillumination images, and so forth. Convoluted neural networks (CNN) demonstrated high sensitivity, specificity, and area under the curve in the caries detection, segmentation, and classification tests. Notably, AI in conjunction with periapical and panoramic radiography images yielded better accuracy in detecting and diagnosing dental caries. CONCLUSION: AI models, especially convolutional neural network (CNN)-based models, have an enormous amount of potential for accurate, objective dental caries diagnosis and detection. However, ethical considerations and cautious adoption remain critical to its successful integration into routine practice.
Asunto(s)
Inteligencia Artificial , Caries Dental , Humanos , Caries Dental/diagnóstico , Caries Dental/diagnóstico por imagen , Redes Neurales de la Computación , Algoritmos , Máquina de Vectores de Soporte , Radiografía Panorámica/métodos , Revisiones Sistemáticas como AsuntoRESUMEN
BACKGROUND: The adoption of convolutional neural networks (CNNs) for decoding electroencephalogram (EEG)-based motor imagery (MI) in brain-computer interfaces has significantly increased recently. The effective extraction of motor imagery features is vital due to the variability among individuals and temporal states. METHODS: This study introduces a novel network architecture, 3D-convolutional neural network-generative adversarial network (3D-CNN-GAN), for decoding both within-session and cross-session motor imagery. Initially, EEG signals were extracted over various time intervals using a sliding window technique, capturing temporal, frequency, and phase features to construct a temporal-frequency-phase feature (TFPF) three-dimensional feature map. Generative adversarial networks (GANs) were then employed to synthesize artificial data, which, when combined with the original datasets, expanded the data capacity and enhanced functional connectivity. Moreover, GANs proved capable of learning and amplifying the brain connectivity patterns present in the existing data, generating more distinctive brain network features. A compact, two-layer 3D-CNN model was subsequently developed to efficiently decode these TFPF features. RESULTS: Taking into account session and individual differences in EEG data, tests were conducted on both the public GigaDB dataset and the SHU laboratory dataset. On the GigaDB dataset, our 3D-CNN and 3D-CNN-GAN models achieved two-class within-session motor imagery accuracies of 76.49% and 77.03%, respectively, demonstrating the algorithm's effectiveness and the improvement provided by data augmentation. Furthermore, on the SHU dataset, the 3D-CNN and 3D-CNN-GAN models yielded two-class within-session motor imagery accuracies of 67.64% and 71.63%, and cross-session motor imagery accuracies of 58.06% and 63.04%, respectively. CONCLUSIONS: The 3D-CNN-GAN algorithm significantly enhances the generalizability of EEG-based motor imagery brain-computer interfaces (BCIs). Additionally, this research offers valuable insights into the potential applications of motor imagery BCIs.
Asunto(s)
Interfaces Cerebro-Computador , Electroencefalografía , Imaginación , Redes Neurales de la Computación , Humanos , Imaginación/fisiología , Adulto , Actividad Motora/fisiología , Encéfalo/fisiología , Procesamiento de Señales Asistido por ComputadorRESUMEN
OBJECTIVE: Despite the prevalent use of the general linear model (GLM) in fMRI data analysis, assuming a pre-defined hemodynamic response function (HRF) for all voxels can lead to reduced reliability and may distort the inferences derived from it. To overcome the necessity of presuming a specific model for the hemodynamic response, we introduce a semi-supervised automatic detection (SAD) method. MATERIALS AND METHODS: The proposed SAD method employs a Bi-LSTM neural network to classify high temporal resolution fMRI data. Network training utilized an fMRI dataset with 75-ms temporal resolution in an iterative scheme. Classification performance was evaluated on a second fMRI dataset from the same participant, collected on a different day. Comparative analysis with the standard GLM approach was conducted to evaluate the cooperative effectiveness of the SAD method. RESULTS: The SAD method performed well based on the classification scores: true-positive rate = 0.961, area under the receiver operating curve = 0.998, true-negative rate = 0.99, F1-score = 0.979, False-negative rate = 0.038, false-discovery rate = 0.002, false-positive rate = 0.002 at 75-ms temporal resolution. CONCLUSION: SAD can detect hemodynamic responses at 75-ms temporal resolution without relying on a specific shape of an HRF. Future work could expand the use cases to include more participants and different fMRI paradigms.
RESUMEN
This paper presents a new hybrid learning and control method that can tune their parameters based on reinforcement learning. In the new proposed method, nonlinear controllers are considered multi-input multi-output functions and then the functions are replaced with SNNs with reinforcement learning algorithms. Dopamine-modulated spike-timing-dependent plasticity (STDP) is used for reinforcement learning and manipulating the synaptic weights between the input and output of neuronal groups (for parameter adjustment). Details of the method are presented and some case studies are done on nonlinear controllers such as Fractional Order PID (FOPID) and Feedback Linearization. The structure and the dynamic equations for learning are presented, and the proposed algorithm is tested on robots and results are compared with other works. Moreover, to demonstrate the effectiveness of SNNFOPID, we conducted rigorous testing on a variety of systems including a two-wheel mobile robot, a double inverted pendulum, and a four-link manipulator robot. The results revealed impressively low errors of 0.01 m, 0.03 rad, and 0.03 rad for each system, respectively. The method is tested on another controller named Feedback Linearization, which provides acceptable results. Results show that the new method has better performance in terms of Integral Absolute Error (IAE) and is highly useful in hardware implementation due to its low energy consumption, high speed, and accuracy. The duration necessary for achieving full and stable proficiency in the control of various robotic systems using SNNFOPD, and SNNFL on an Asus Core i5 system within Simulink's Simscape environment is as follows: - Two-link robot manipulator with SNNFOPID: 19.85656 hours - Two-link robot manipulator with SNNFL: 0.45828 hours - Double inverted pendulum with SNNFOPID: 3.455 hours - Mobile robot with SNNFOPID: 3.71948 hours - Four-link robot manipulator with SNNFOPID: 16.6789 hours. This method can be generalized to other controllers and systems like robots.
RESUMEN
The compressive strength of concrete depends on various factors. Since these parameters can be in a relatively wide range, it is difficult for predicting the behavior of concrete. Therefore, to solve this problem, an advanced modeling is needed. The aim of the literature is to achieve an ideal and flexible solution for predicting the behavior of concrete. Therefore, it is necessary to develop new approaches. Artificial Neural Networks (ANNs) have evolved from a theoretical method to a widely utilized technology by successful applications for a variety of issues. Actually, ANNs are a strong computing tool that provides the right solutions to problems that are difficult to use conventional methods. Inspired by the biological neural system, these networks are now widely used for solving a wide range of complicated problems in civil engineering. This study''s target is evaluating the performance of developed African vulture optimization algorithm (DAVOA)-Elman neural networks (ENNs) by considering different input parameters in predicting the self-compacting concrete compressive strength. Hence, once 8 parameters and again to get as close as possible to the prediction conditions in the laboratory, 140 parameters entered to the improved version of Elman Neural Networks as input. According to the results, the element network has the lowest mean squares of the test error in predicting the compressive strength of 7 and 28 days in 100 repetitions. Further, in predicting both compressive strengths, the element grid with the Logsig-Purelin interlayer transfer function has the lowest test error, which determines the optimal transfer function. Moreover, the results showed that DAVOA as a reliable tool with time and cost savings have high power in predicting the desired characteristics. Also, in predicting both 7-day and 28-day compressive strength, networks built with 140 parameters have a 74.54 and 70.44% improvement in test error over 8-parameter networks, respectively, which directly affects this effect. Further parameters are considered as input to the network error rate in predicting the desired properties.
RESUMEN
The use of convolutional neural networks can revolutionize XRD analysis by significantly reducing processing times. Demonstration against synthetic and real mineral mixture data provide a first assessment of the accuracy of such methods.