Epilepsy is a neurological disorder that affects millions of people worldwide, necessitating reliable non-invasive prediction methods to improve patient outcomes. Although Electroencephalogram (EEG) signals remain the primary standard, recent advances have introduced Electrocardiogram (ECG) signals, specifically Heart Rate Variability (HRV), as a promising, accessible seizure biomarker. Unlike previous reviews, which have focused primarily on clinical perspectives, this study extends the focus and presents a systematic evaluation of machine learning methodologies applied to ECG-based seizure prediction for real-world implementation. In our analysis, two primary modeling paradigms are distinguished: classification, which discriminates between seizure states, and anomaly detection, which identifies deviations from non-seizure baselines. Our analysis reveals that while advanced machine learning models demonstrate superior sensitivity, they are frequently constrained by data scarcity and lack of interpretability. Furthermore, we identify that morphological and non-linear features can provide higher discriminative power than traditional metrics, although they are underutilized. This review consolidates current findings on public datasets, signal preprocessing, feature extraction, and validation protocols, while identifying key barriers such as physiological non-specificity. We conclude by proposing a strategic roadmap for future research, emphasizing the integration of multimodal sensor fusion and explainable design to enhance the robustness, scalability, and clinical trustworthiness of predictive systems.

Reliable coronary artery segmentation from computed tomography angiography (CTA) is of great importance for the diagnosis and treatment planning of the coronary artery disease (CAD). In this paper, we propose a new dual dense connectivity architecture, designed based on the popular UNETR model, which is not only superior in learning multi-scale features but also beneficial to explicitly capture boundaries, through intra-block dense feature fusion and dense multi-scale skip connections. Our model process 2D CTA slices and incorporates both global self-attention and local depth-wise convolutional attention in a hybrid transformer block. Extensive validation on two independent datasets, ImageCAS and ASOCA, shows excellent state-of-the-art performance (Dice: 0.900 (internal validation) and 0.89 (external validation)) across all metrics on existing benchmarks. Ablation experiments validate the role of each architectural element as well as edge-aware composite loss. The presented method provides a robust and practical solution for accurate coronary artery segmentation, having promising impact on automated diagnostic pipeline as well as intervention planning.

Human action recognition is one of the most challenging tasks in machine intelligence societies. It is important to extract discriminative spatial-temporal features to learn action representation. However, the discriminative information of videos is usually sparse and mixed with a large amount of redundant and interference information, which results in poor performance and recognition failure. Spatial temporal Attention modules enable the network to learn discriminative feature representation of different human actions. One critical key issue which is often missed in the design of these modules is visual tempo of actions. Since a video is formed by a set of spatial changes over time, in this paper, a visual tempo based spatial-temporal attention mechanism is proposed which helps to focus the model on the most meaningful changes in space and time. The proposed attention module is able to flexibly integrated into recurrent networks in a plug-and-play manner. Experimental results on UCF101, HMDB51, and Kinetics-400 demonstrate that the proposed model achieves superior performance among RCNN-based architectures and remains highly competitive with recent state-of-the-art methods, effectively balancing high accuracy with computational efficiency.

Drug discovery requires a tremendous amount of time and cost. Computational drug-target interaction prediction, a significant part of this process, can reduce these requirements by narrowing the search space for wet lab experiments. In this survey, we provide comprehensive details of graph machine learning-based methods in predicting drug-target interaction, as they have shown promising results in this field. These details include the overall framework, main contribution, datasets, and their source codes. The selected papers were mainly published from 2020 to 2024. Prior to discussing papers, we briefly introduce the datasets commonly used with these methods and measurements to assess their performance. Finally, future challenges and some crucial areas that need to be explored are discussed.

Link prediction in temporal and multiplex networks is a crucial issue across both applied and scientific disciplines within the study of complex networks. Recent advances in hardware and the increased availability of computational resources have enhanced our capacity to tackle this problem more effectively.Link prediction in multiplex and temporal networks faces challenges such as inter-layer dependencies and the temporal expansion and contraction of the network. This paper introduces MetaLink, a novel approach designed for link prediction within such temporal multiplex networks. MetaLink leverages knowledge obtained from various temporal network snapshots by employing two innovative methods for subsequent temporal snapshot. It efficiently facilitates the transfer of knowledge across different temporal snapshots. The intra-layer knowledge transfer is governed by a time-decay function, while inter-layer knowledge is learned in a step wise and transferred using the Model-Agnostic Meta-Learning (MAML) algorithm from one snapshot (task) to another. Our findings demonstrate that MetaLink significantly outperforms static single-layer and multiplex methods, showing improvements of 2 to 5 percent, and exhibits up to a 3 percent enhancement over existing temporal methods.

Convergent evidence has suggested a significant effect of bumetanide exposure on brain connectivity in patients with autism spectrum disorder (ASD), yet the characteristics of favorable treatment outcome remains largely unknown. In this work, we aimed to elucidate the primary group distinctions in the temporal pattern of dynamic functional connectivity (dFC) estimated from task based EEG among individuals with ASD and healthy controls (HC) and explore the effect of bumetanide on brain network states. Through a systematic analysis, we investigated the ASD-related intrinsic alterations of dFC at baseline, followed by a longitudinal study to examine the influence of bumetanide treatment on these abnormalities by comparing patients at baseline and follow-up. EEG data was recorded from 9 individuals with ASD and 9 HC at baseline. After 3 months of bumetanide treatment, EEG data was re-recorded from the individuals with ASD. Utilizing phase locking value, sliding-window analysis, and K-means clustering, we obtained dFC for the alpha, beta, and gamma bands, and observed three repeating brain states. Two temporal metrics of connectivity state expression including fraction of time (FT) and dwell time (DT) were calculated. The results showed that in the alpha and gamma bands, FT and DT in ASD groups before bumetanide treatment were higher in the strong connection state and lower in the weak connection state compared to the HC group. Also, the use of bumetanide led to a significant decrease in the difference between the HC and ASD groups, although the adjustment made in the gamma band was more significant.

Profit maximization and risk mitigation require good financial market predictions. Financial markets have a correlated nature, which means that there are some shared patterns between them; therefore, learning about one market might help understand the behavior of others. End-to-end training techniques have proven successful in financial markets, but they have flaws, such as picking up noise and failing to account for the complicated relationships across markets. We present a promising model for predicting financial markets using the correlation between the two markets, which draws inspiration from the recent progress in disentanglement learning. This model learns to disentangle representations of features shared between markets from specific representations, and removes features that cause interference. We utilized a dilated convolutional neural network as an encoder to extract features while using self-attention and cross-attention to capture specifics and shared patterns. Our model uses Dynamic Time Warping (DTW) to minimize the similarity between specific and shared patterns. It also combines DTW's alignment-based similarity with the Mean Square Error (MSE) to determine the optimal balance between alignment and prediction accuracy. We conducted our experiments using datasets that included the closing prices of Apple, Samsung, Bitcoin, Ethereum, Meta platforms, and the X platform. Spearman's rank correlation coefficient was used to evaluate the disentanglement by describing the relationship between the extracted representations. The findings confirm that our model surpasses state-of-the-art approaches in prediction error, financial risk assessment, correlation evolution, and prediction net curves, thereby giving market participants more trust in their decisions.

The increasing popularity of ride-hailing services has led to a growing need for optimal utilization of resources and minimization of passenger wait times. A major prerequisite for this goal is to develop accurate taxi demand prediction models. In this study, we propose a novel approach for district-centric online taxi demand prediction. We first used historical data and Pearson coefficient to identify correlated districts. In order to dynamically weight the importance of related districts, we utilized an architecture which is based on attention mechanism. Next, for demand prediction in each district we used an LSTM network and finally to refine the prediction models we used the Gradient Boosting algorithm. To evaluate the effectiveness of our approach, we leverage the real data of Snapp!, a ride-hailing service in Iran. Results confirm that the proposed model outperforms the state-of-the-art methods in terms of prediction error.

BackgroundDental caries, also known as tooth decay, is a widespread and long-standing condition that affects people of all ages. This ailment is caused by bacteria that attach themselves to teeth and break down sugars, creating acid that gradually wears away at the tooth structure. Tooth discoloration, pain, and sensitivity to hot or cold foods and drinks are common symptoms of tooth decay. Although this condition is prevalent among all age groups, it is especially prevalent in children with baby teeth. Early diagnosis of dental caries is critical to preventing further decay and avoiding costly tooth repairs. Currently, dentists employ a time-consuming and repetitive process of manually marking tooth lesions after conducting radiographic exams. However, with the rapid development of artificial intelligence in medical imaging research, there is a chance to improve the accuracy and efficiency of dental diagnosis.MethodsThis study introduces a data-driven model for accurately diagnosing dental decay through the use of Bitewing radiology images using convolutional neural networks. The dataset utilized in this research includes 713 patient images obtained from the Samin Maxillofacial Radiology Center located in Tehran, Iran. The images were captured between June 2020 and January 2022 and underwent processing via four distinct Convolutional Neural Networks. The images were resized to 100 x 100 and then divided into two groups: 70% (4219) for training and 30% (1813) for testing. The four networks employed in this study were AlexNet, ResNet50, VGG16, and VGG19.ResultsAmong different well-known CNN architectures compared in this study, the VGG19 model was found to be the most accurate, with a 93.93% accuracy.ConclusionThis promising result indicates the potential for developing an automatic AI-based dental caries diagnostic model from Bitewing images. It has the potential to serve patients or dentists as a mobile app or cloud-based diagnosis service (clinical decision support system).

Prostate cancer is one of the most common and fatal diseases among men, and its early diagnosis can have a significant impact on the treatment process and prevent mortality. Since it does not have apparent clinical symptoms in the early stages, it is difficult to diagnose. In addition, the disagreement of experts in the analysis of magnetic resonance images is also a significant challenge. In recent years, various research has shown that deep learning, especially convolutional neural networks, has appeared successfully in machine vision (especially in medical image analysis). In this research, a deep learning approach was used on multi-parameter magnetic resonance images, and the synergistic effect of clinical and pathological data on the accuracy of the model was investigated. The data were collected from Trita Hospital in Tehran, which included 343 patients (data augmentation and learning transfer methods were used during the process). In the designed model, four different types of images are analyzed with four separate ResNet50 deep convolutional networks, and their extracted features are transferred to a fully connected neural network and combined with clinical and pathological features. In the model without clinical and pathological data, the maximum accuracy reached 88%, but by adding these data, the accuracy increased to 96%, which shows the significant impact of clinical and pathological data on the accuracy of diagnosis.

Emerging evidence suggests that cognitive impairment associated with brain network disorders in people with autism could be improved with medications such as bumetanide. However, the extent to which bumetanide is effective in improving brain function in these individuals has not been adequately studied. The main purpose of this study is to investigate the nonlinear brain connectivity and topological changes in brain networks of people with autism spectrum disorders (ASD) after a three-month course of bumetanide treatment. We used electroencephalography (EEG) data of nine participants recorded during the face emotion recognition activity in two stages before and after bumetanide treatment. Brain connectivity matrix was calculated using a neural network-based estimator. Graph criteria and statistical tests have been used to determine the effects of bumetanide treatment on children and adolescents with autism. Bumetanide treatment significantly alters the brain connectivity networks based on stimuli type. Differences in brain connectivity related to the sad stimuli are more significant. The most of the significant changes of the strength graph metric was in the occipital electrodes and electrodes related to the right hemisphere. These results suggest that bumetanide may affect effective connectivity and be used a promising treatment for improving social interactions in patients with autism. It also suggests that brain connectivity patterns can be considered as a neural marker to be used in the development of new therapies.

Dental caries is one of the oral health problems and the most common chronic infectious disease of childhood, and neural networks and artificial intelligence are increasingly being used in the field of dentistry. This review study aims to review studies published in the field of artificial intelligence and neural networks and dentistry. A search for studies in four databases, including Springer, ScienceDirect, PubMed (MedLine), and Institute of Electrical and Electronics Engineers (IEEE) was done. Finally, 28 studies were reviewed, most of which used Bitewing and Periapical images for the classification and detection of dental caries. The image databases ranged from 55 to 3000 and several evaluation metrics were used in the selected studies. The research questions were designed and reviewed based on PICOS (P stands for patient or problem, I stands for intervention, C stands for control or comparison, and O stands for outcomes). The majority of the studies also used pre-processing and data augmentation methods. The diversity between the networks used and the output evaluation criteria have made direct research comparisons challenging. The main focus of this research was on caries detection using deep learning methods and neural networks, especially convolutional neural networks that are suitable for images. The traditional methods of detecting caries, other than the methods based on artificial intelligence, have not been investigated in this research. Also, the main caries were interproximal and proximal caries in molars and premolars. The main difference between this and previous works is the use of more up-to-date articles (2016 to 2023) studies with an organized manner of reviewing, which is based on the types of images used.

In this article, we proposed a plan based on Adaptive Elastic-net Sliced Inverse Regression to identify risk factors for the coronavirus disease (Covid-19) disease in the presence of collinearity between explanatory variables. Considering the penalty of elastic-net and sliced inverse regression, this method leads to sufficient dimension reduction and the presentation of a more stable and accurate model for variable selection.We applied the proposed method to simulated data and a new real-world Covid-19 disease dataset. We observed that the proposed method reduced the experimental standard error of bootstrapping by 12\% and 13\% compared to the previous superior methods in this approach, respectively, for both datasets. According to the results, during the outbreak of the Covid disease and its re-intensification, countries should quickly implement the following policies: declaring quarantine with minimal exceptions, making vaccines available by prioritizing specific groups, declaring a ban on gatherings, especially gatherings of more than 1000 people, closing schools at all levels, closing some works or declaring remote work, and holding information campaigns. Especially countries with more 0-14-year-old population, higher life expectancy, lower human development index, and colder weather should make more serious decisions in their implementation because they are more at risk.

Building detection and change detection using remote sensing images can help urban and rescue planning. Moreover, they can be used for building damage assessment after natural disasters. Currently, most of the existing models for building detection use only one image (pre-disaster image) to detect buildings. This is based on the idea that post-disaster images reduce the model's performance because of presence of destroyed buildings. In this paper, we propose a siamese model, called SiamixFormer, which uses pre- and post-disaster images as input. Our model has two encoders and has a hierarchical transformer architecture. The output of each stage in both encoders is given to a temporal transformer for feature fusion in a way that query is generated from pre-disaster images and (key, value) is generated from post-disaster images. To this end, temporal features are also considered in feature fusion. Another advantage of using temporal transformers in feature fusion is that they can better maintain large receptive fields generated by transformer encoders compared with CNNs. Finally, the output of the temporal transformer is given to a simple MLP decoder at each stage. The SiamixFormer model is evaluated on xBD, and WHU datasets, for building detection and on LEVIR-CD and CDD datasets for change detection and could outperform the state-of-the-art.

Every facial expression involves one or more facial action units appearing on the face. Therefore, action unit recognition is commonly used to enhance facial expression detection performance. It is important to identify subtle changes in face when particular action units occur. In this paper, we propose an architecture that employs local features extracted from specific regions of face while using global features taken from the whole face. To this end, we combine the SPPNet and FPN modules to architect an end-to-end network for facial action unit recognition. First, different predefined regions of face are detected. Next, the SPPNet module captures deformations in the detected regions. The SPPNet module focuses on each region separately and can not take into account possible changes in the other areas of the face. In parallel, the FPN module finds global features related to each of the facial regions. By combining the two modules, the proposed architecture is able to capture both local and global facial features and enhance the performance of action unit recognition task. Experimental results on DISFA dataset demonstrate the effectiveness of our method.

Accurate prediction of financial market trends can have a great impact on maximizing profits and avoiding risks. Conventional methods, e.g., regression or SVR, or end-to-end training approaches, coined as deep learning algorithms, have restraints as a consequence of capturing noisy and unnecessary data. Financial market’s data are composed of stock’s price time series that are correlated, and each time series has both global and local dynamics. Inspired by recent advancements in disentanglement representation learning, in this paper, we present a promising model for predicting financial markets that learn disentangled representations of features and eliminate those features that cause interference. Our model uses the informer encoder to extract features, capturing global–local patterns by using the time and frequency domains, augmenting the clean features with time and frequency-based features, and using the decoder to predict. To be more specific, we adopt contrastive learning in the time and frequency domains to learn both global and local patterns. We argue that our methodology, disentangling and learning the influential factors, holds the potential for more accurate predictions and a better understanding of how time series move and behave. We conducted our experiments using the S&P 500, CSI 300, Hang Seng, and Nikkei 225 stock market datasets to predict their next-day closing prices. The results showed that our model outperformed existing methods in terms of prediction error (mean squared error and mean absolute error), financial risk measurement (volatility and max drawdown), and prediction net curves, which means that it may enhance traders’ profits.

In this paper, we propose a transfer learning-based approach for road sign classification using pre-trained CNN models. We evaluate the performance of our fine-tuned VGG-16, VGG-19, ResNet50 and EfficientNetB0 models on the German Traffic Sign Recognition Benchmark (GTSRB) test dataset. Our work makes several contributions, including the utilization of transfer learning with pre-trained CNN models, the integration of augmentation techniques, and a comprehensive evaluation of multiple state-of-the-art architectures on the road sign classification task. Our experimental results show that the proposed approach achieves an average accuracy of 99.2

The early diagnosis of subjects with mild cognitive impairment (MCI) is an effective appliance of prognosis of Alzheimer’s disease (AD). Electroencephalogram (EEG) has many advantages compared to other methods in the analysis of AD in an early stage. In this paper, two different deep learning (DL) architectures, including modified convolutional (CNN) and convolutional autoencoder (Conv-AE) neural networks (NNs), are proposed for classifying subjects into AD, mild cognitive impairment (MCI), and healthy control (HC) data based on scalp EEG recordings. The database includes 19-channel EEG recorded from 61 healthy control, 56 MCI, and 63 AD subjects. Time–frequency representation (TFR) is used to extract desirable features from EEG signals. Continuous wavelet transform (CWT) with Mexican hat function (MHf) as its mother wavelet is used for the selected TFR. The average accuracy obtained for the modified convolutional network and the convolutional auto-encoder network are 92% and 89%, respectively. The proposed networks in this study have superiority over those in similar studies not only by providing 10% increase in classification accuracy but also by improving the number of classes for similar data. In addition, the obtained accuracy of our networks was significantly higher than that of conventional machine learning methods. We believe the results illustrate DL architectures to be a good tool to handle EEG analysis, because of the ability to deal directly with inaccurate, inconsistent, and Para complete data, thereby providing a practical analysis.

Multiplex networks are very flexible at showing heterogeneous relationships between identical entities. Link prediction is a fundamental problem in network science. There are many studies on link prediction in complex networks, but few studies were conducted on link prediction in multiplex networks. This study proposes a method for estimating link likelihood in multiplex networks based on the Node-Accessibility-Distribution (NAD) and the co-evolving factors of layers. The NAD is introduced as a probabilistic measure to find local and pseudo-global structural features of nodes in layers of the multiplex network. The probabilistic distance among nodes is calculated using Jensen–Shannon diversity. Since the evolution of one layer subsequently affects the dynamics of other layers, this study introduces the co-evolving factors as criteria for determining the effect of the evolution of layers in the formation of new links in the target layer. In order to estimate the co-evolving factors, logistics regression and Maximum Likelihood Estimation(MLE) are employed. The proposed method is evaluated with six real-world datasets. The results show that the proposed approach has a better average AUC and precision than the state-of-the-art methods. Based on various datasets, the AUC and precision were improved by 1% to 5% compared with the state-of-the-art.