In light of the growing prevalence of the autoimmune disease multiple sclerosis (MS), accurate detection of MS lesions in brain magnetic resonance imaging (MRI) images plays a critical role in assisting neurologists with timely diagnosis. The high similarity between MS lesions and normal brain tissues, however, makes this task particularly challenging. Although numerous deep-learning-based approaches have been proposed for the automatic segmentation of MS lesions, the method presented in this study has achieved superior results. ZechariahNet is a U-Net-based architecture that integrates transition down blocks, squeeze-attention (SA) blocks, dense blocks, and Convolutional LSTM (C-LSTM) blocks within a 3D CNN framework. By jointly exploiting spatial-temporal information from three consecutive MRI slices (previous, current, and subsequent) and strategically applying C-LSTM modules across the encoder and decoder paths, the proposed model effectively captures the neighborhood dependencies for enhanced feature extraction and reconstruction. These architectural innovations significantly improve segmentation accuracy, enabling ZechariahNet to achieve a dice similarity coefficient (DSC) of 84.72%, outperforming existing state-of-the-art methods.

Accurate and automatic segmentation of brain tumors from Magnetic Resonance Imaging (MRI) plays a vital role in diagnosis, treatment planning, and disease monitoring. While Convolutional Neural Network (CNN)-based architectures excel at extracting local features, they are limited in comprehending global image context; conversely, Transformer models are superior in modeling long-range and global dependencies, thereby addressing this CNN limitation. In this paper, we propose a novel hybrid U-shaped architecture that effectively combines the strengths of both approaches by utilizing a pre-trained Swin-Transformer backbone as the encoder to extract hierarchical and context-rich global features. The key innovation is the introduction of two sophisticated spatial attention modules to refine and adapt the encoder features specifically for the medical domain, along with a channel attention-aided upsampling module in the decoder to adaptively and optimally re-weight the information received from the skip connections. Evaluations conducted on the challenging BRISC dataset show that our proposed method outperforms previous state-of-the-art models, achieving an 80.6% score in Weighted IoU and 88.6% in the Dice Coefficient, thereby proving the efficiency of combining the Transformer with dual attention mechanisms.

Multiple Sclerosis (MS) and Myelitis are serious inflammatory spinal cord disorders with overlapping clinical symptoms and radiological characteristics, making accurate differentiation challenging yet clinically essential. Early and precise diagnosis is critical for guiding treatment strategies and improving patient outcomes. In this study, we propose KhayyamNet, a novel hybrid deep learning architecture designed to fuse complementary local and global representations for the accurate diagnosis of MS and Myelitis using spinal MRI. To improve robustness and generalization capability, a comprehensive preprocessing strategy including data augmentation and intensity normalization is also applied to reduce noise and address data variability. The proposed architecture combines three complementary deep learning models for feature extraction composed of Xception for high-level semantic features, Convolutional Neural Networks (CNNs) for fine-grained local patterns, and Vision Transformers (ViTs) for global contextual representations via attention mechanisms. Extracted features are then fused and refined using the Minimum Redundancy Maximum Relevance (MRMR) algorithm to eliminate redundancy and retain the most informative signals. Finally, a Random Forest (RF) classifier utilizes the optimized feature set to achieve accurate and robust differentiation between MS, Myelitis, and control spinal MRIs. Experimental results demonstrate that KhayyamNet outperforms existing methods by achieving an average classification accuracy of 98.15±0.80%. This framework demonstrates promising performance for the automated analysis of spinal MRIs and shows potential to assist in the differentiation of MS and Myelitis. While these findings highlight the potential of KhayyamNet for automated MRI interpretation, its evaluation is limited to a single-center dataset, and further validation on external multi-center data is required.

Anxiety disorders are a significant challenge to global health, yet current diagnostic methods rely mainly on subjective and episodic assessment. The application of artificial intelligence (AI) in combination with neural and physiological signals is a promising pathway toward objective and continuous monitoring. This review provides a comprehensive evaluation of machine learning (ML) and deep learning (DL) techniques applied to electroencephalography (EEG), electrocardiography (ECG), Photoplethysmography (PPG), and electrodermal activity (EDA) for anxiety detection. A systematic review of the literature covering the period 2015-2025 indicates that the field has evolved from traditional ML systems based on hand-crafted features to modern end-to-end deep learning schemes. Our review indicates that while classical models remain effective, hybrid models such as CNN-LSTM and more sophisticated architectures, like Transformers, consistently deliver state-of-the-art results, particularly in multimodal data integration scenarios. Despite the impressive accuracies reported, a critical examination identifies several key challenges including reliance on narrow, laboratory-trained datasets, a lack of standardized validation procedures, and limited transparency in complex models, collectively impeding clinical translation. Progress in this field requires the creation of large-scale, clinically-validated datasets; the development of fault-tolerant, generalizable, and interpretable schemes; and a transition from basic state classification towards longitudinal, personalized monitoring in support of just-in-time intervention. As a comprehensive and up-to-date narrative synthesis of AI-driven anxiety detection using neural and physiological signals, this review consolidates a fragmented literature, articulates the translational gaps limiting real-world deployment, and establishes a clear research agenda for the development of clinically impactful digital mental health technologies.

Anxiety can significantly impact individuals’ daily lives, and can manifest at varying levels from mild to severe. Traditionally, psychologists and psychiatrists assess anxiety primarily through self-report questionnaires. However, advances in computer-aided technologies and neuroimaging techniques offer promising tools to enhance diagnostic accuracy. In this study, we propose a novel deep learning model designed to extract frequency-based features from electroencephalogram (EEG) signals which provide insights into the neural patterns associated with anxiety. Our model consists of a convolutional neural network (CNN), a multi-head attention transformer, and an attention module to effectively capture EEG features distinguishing normal and anxious states. We validated our approach using a publicly available EEG dataset called DASPS, collected from 23 participants, where self-reported anxiety levels were categorized into normal and anxious conditions. The anxious condition was further subdivided into four levels of anxiety based on its severity. The proposed model achieved classification 82.94% accuracy for binary classification (normal vs. anxious) and 74.05% average accuracy for multi-class classification (normal, mild, moderate, and severe anxiety). These results highlight the effectiveness of our approach in leveraging EEG-based frequency features for anxiety assessment across different levels of severity.

This study aims to characterize and compare the functional neural networks associated with different olfactory stimuli, including air, non-social odours, and human body odours. We introduce a novel processing pipeline based on event-related functional magnetic resonance imaging (fMRI) and graph theory for network identification. To ensure the stability and small worldness of the characterized networks, we conduct statistical validations, network modularity assessments, and robustness measurement against local attacks. The key hypothesis is that human body odours (so-called social odours) and non-social odours engage distinct neural networks, particularly in regions responsible for social processing. We found that the posterior medial orbitofrontal cortex (pmOFC) and fusiform face area (FFA) demonstrate stronger centrality in the body odour network than the non-social odour and air networks. This observation supports the idea that social and olfactory information are integrated in the body odour network. Additionally, the anterior insula (INSa), posterior piriform cortex (PPC), and amygdala (AMY) exhibit high influence in air and odour networks by achieving higher centrality indices and playing a major role in improving the global efficiency. These findings offer impactful insight into how air, non-social, and social odours recruit distinct neural circuits, reinforcing the role of olfaction in human social behavior.

Anxiety can have a profound effect on our lives. People may experience different levels of anxiety ranging from mild to severe, and psychologists and psychiatrists mainly rely on self-report questionnaires to measure this. However, new computer-aided technologies and neuroimaging techniques could significantly help them to verify their diagnosis. In this paper, a novel deep learning model is designed to precisely screen electroencephalogram (EEG) signals to characterise the neural patterns associated with anxiety. Our deep learning model integrates a convolutional neural network, an attention module and a recurrent neural network to effectively estimate the EEG features signifying normal and anxious emotions. In order to improve the performance of the model, we adopted a data transformation approach to generate a spatio-temporal representation of EEG data. We evaluated the performance of our proposed model using a publicly available EEG data set acquired from 23 subjects who reported feeling normal or anxious. Anxiety was further categorised into four sub-groups based on the level of anxiety. The model achieved a classification accuracy of 94.24% and 92.58% for binary (i.e normal and anxious) and multi-class (i.e normal, light, moderate and severe anxiety) scenarios, respectively. The obtained results indicated the success of our proposed model in learning EEG patterns across various levels of anxiety. Additionally, comparing the obtained results with previously published studies demonstrated considerable superiority of our method.

Human sensory systems are capable of encoding emotion-related signals during social interactions (e.g., fearful or happy facial expressions). In this regard, many emotion elicitation procedures have been reported within the scope of social signal processing research. Most of these procedures rely on socially relevant stimuli, such as emotional face images, videos, and, more recently, virtual reality (VR) scenes. Unfortunately, procedures involving cross-modal interactions beyond visual and acoustic stimuli, such as olfaction, are still scarce. In this sense, neuroscience supports a close link between the olfactory and affective systems. Moreover, experimental research has reported faster appraisals of emotional face images when congruent valence-laden artificial scents were presented (e.g., positive scent-happy face; negative scent-fearful face). Interestingly, recent findings indicate that emotion-related human-body odors (HBOs) might also modulate affective appraisals during a neutral virtual reality experience. However, whether and how emotion-related HBOs modulate affective VR experiences requires further examination. Here, an approach to this research question is proposed from a Virtual Reality-based Behavioral Biomarker (VRBB) experimental framework. Concretely, in the first place, a novel affective elicitation procedure based on social-emotional VR is introduced, wherein electro-dermal activity (EDA), heart-rate variability (HRV), electroencephalography (EEG), and affective appraisals, will be accounted for. In a second step, the modulating role of HBOs will be investigated regarding those measures. This work presents the envisioned model, details of the devised VEs, and a research design to test concrete hypotheses.

Facial emotion processing by the brain plays a decisive role in human social interactions. This signal helps us interpret and predict people's behaviours. However, other social signals such as human voices or human body odours may facilitate or impair the identification of facial expressions. Here we studied the effects of emotional human body odours on face processing by measuring evoked neural responses and brain connectivity using the electroencephalogram (EEG). We used an emotion recognition task in which the participants attributed an emotion (i.e. happy vs fearful) to a presented face image while simultaneously exposed to emotional body odours. First, we measured face related potentials (FRP)s including P100 and N170 components. Statistical analyses revealed significant differences among FRPs recorded in different odour conditions. Second, we used a hierarchical Bayesian approach including a group dynamic causal model (DCM) followed by parametric empirical Bayes (PEB) to characterize the brain network explaining differences between FRPs. Our preliminary results suggested that different brain networks contribute to neutral face processing in the presence of different emotional body odours.

In this paper, a new technique based on the manipulation of stator current signature for induction machines fault diagnosis is introduced. The goal of the proposed method is to demodulate the characteristic frequencies from supply frequency and preserve the information of the supply frequency and its harmonics. The proposed method can be easily implemented and used in experimental systems due to its low computational complexity. The validity of the proposed method is proved through theoretical analysis and experimental results in steady-state and transient conditions. In this regard, the 270-W wound rotor induction machine (WRIM) is tested under different fault severities and load levels.

Dictionary learning was introduced for sparse image representation. Today, it is a cornerstone of image classification. We propose a novel dictionary learning method to recognise images of handwritten numbers. Our focus is to maximise the sparse-representation and discrimination power of the class-specific dictionaries. We, for the first time, adopt a new feature space, i.e., histogram of oriented gradients (HOG), to generate dictionary columns (atoms). The HOG features robustly describe fine details of hand-writings. We design an objective function followed by a minimisation technique to simultaneously incorporate these features. The proposed cost function benefits from a novel class-label penalty term constraining the associated minimisation approach to obtain class-specific dictionaries. The results of applying the proposed method on various handwritten image databases in three different languages show enhanced classification performance (∼98%) compared to other relevant methods. Moreover, we show that combination of HOG features with dictionary learning enhances the accuracy by 11% compared to when raw data are used. Finally, we demonstrate that our proposed approach achieves comparable results to that of existing deep learning models under the same experimental conditions but with a fraction of parameters.

One of the major challenges of diagnosing rotor symmetry faults in induction machines (IMs) is severe modulation of fault and supply frequency components. In particular, existing techniques are not able to identify fault components in the case of low slips. In this article, this problem is tackled by proposing a novel approach. First, a new use of singular spectrum analysis (SSA), as a powerful spectrum analyzer, is introduced for fault detection. Our idea is to treat the stator current signature of the wound rotor IM as a time series. In this approach, the current signature is decomposed into several eigenvalue spectra (rather than frequency spectra) to find a subspace where the fault component is recognizable. Subsequently, the fault component is detected using some data-driven filters constructed with the knowledge about characteristics of supply and fault components. Then, an inexpensive peak localization procedure is applied to the power spectrum of the fault component to identify the exact frequency of the fault. The fault detection and localization methods are then combined in a recursive regime to further improve the diagnosis' performance particularly at high rotor speeds and small rotor faults. The proposed approach is data-driven and is directly applied to the raw signal with no suppression or filtration of the frequency harmonics with a low computational complexity. The numerical results obtained with real data at several rotation speeds and fault severities unveil the effectiveness and real-time feature of the proposed approach.

In this paper, the problem of monitoring the electrical pulse of the fetal heart using non-invasive electrocardiogram (ECG) signal is studied. To overcome challenges like the overlap between maternal and fetal heart components and noise in the mother's abdomen signal, we propose a new single-channel signal processing technique to detect the fetal QRS (fQRS) events. We first propose a modified sample entropy technique to filter out noisy and complex signal components. Then, common spatial pattern (CSP) technique is employed to find a new sub-space with reduced overlap among maternal and fetal patterns. Then, a one-dimensional convolutional neural network is trained to detect the fQRS events from the abdominal ECG. The obtained results on several well-established datasets are promising and confirm the effectiveness of the proposed approach.

Deep learning methods have become the key ingredient in the field of computer vision; in particular, convolutional neural networks (CNNs). Appropriating the network architecture and data pre-processing have significant impact on performance. This paper focuses on the classification of handwritten Chinese numbers. Firstly, we applied various methods of pre-processing to our collected image dataset. Secondly, we customised a CNN-based architecture with minimal number of layers and parameters specifically for the task. Experimental results showed that our proposed methods provides superior classification rate of 99.1%. Our results also show that the proposed method has competitive performance compared to smaller neural networks with fewer parameters, e.g. Squeezenet and deeper networks with a larger size and number of parameters, e.g., pre-trained GoogLeNet and MobileNetV2.

Brain-computer interface (BCI) is a powerful system for communicating between the brain and outside world. Traditional BCI systems work based on electroencephalogram (EEG) signals only. Recently, researchers have used a combination of EEG signals with other signals to improve the performance of BCI systems. Among these signals, the combination of EEG with functional near-infrared spectroscopy (fNIRS) has achieved favourable results. In most studies, only EEGs or fNIRs have been considered as chain-like sequences, and do not consider complex correlations between adjacent signals, neither in time nor channel location. In this study, a deep neural network model has been introduced to identify the exact objectives of the human brain by introducing temporal and spatial features. The proposed model incorporates the spatial relationship between EEG and fNIRS signals. This could be implemented by transforming the sequences of these chain-like signals into hierarchical three-rank tensors. The tests show that the proposed model has a precision of 99.6%.

Chemosensory communication is known as an effective way to influence the human emotion system. Phenomena like food selection or motivation, based on chemical signals, present a unique pathway between chemosensory and emotion systems. Human chemosignals (i.e. sweat) which are produced during different emotional states contain associated distinctive odors and are able to induce same emotions in other people. For instance, sweat is known as a social chemosignal participating in social interaction. Chemosignal perception engages a distributed neural network which has not been well characterized yet. In this paper, we use functional magnetic resonance imaging (fMRI) to investigate the neural circuits underlying social emotional chemosignal processing. Chemosignals associated with disgust and neutral conditions were used to induce specific emotional states in fMRI participants during a healthy food judgement. We performed fMRI analysis with the aim of detecting active areas in the brain, followed by a dynamic causal modeling (DCM) analysis. fMRI analysis revealed functional activity in the fusiform face area (FFA), amygdala (AMG) and orbitofrontal cortex (OFC). In order to determine the effective connectivity among these regions as a result of emotional chemosignal processing, a set of dynamic causal models is proposed. Estimating parameters of the proposed models shows that social chemosignals modulate the connections between FFA, AMG and OFC. The results indicate that social chemosignals of disgust converge on orbitofrontal cortex - an area which is a critical region for object appraisal and valuation - after first influencing fusiform face area and amygdala.

Dictionary learning is a cornerstone of image classification. We set out to address a longstanding challenge in using dictionary learning for classification; that is to simultaneously maximise the discriminability and sparse-representability power of the learned dictionaries. Upon this premise, we designed class-specific dictionaries incorporating three factors: discriminability, sparsity and classification error. We integrated these metrics into a unified cost function and adopted a new feature space, i.e., histogram of oriented gradients (HOG), to generate the dictionary atoms. The rationale of using HOG features for designing the dictionaries is their strength in describing fine details of crowded images. The results of applying the proposed method in the classification of Chinese handwritten numbers demonstrated enhanced classification performance $(\sim98\%)$ compared to state-of-the-art deep learning techniques (i.e., SqueezeNet, GoogLeNet and MobileNetV2), but with a fraction of parameters. Furthermore, combination of the HOG features with dictionary learning enhances the accuracy by $11\%$ compared to the case where only pixel domain data are used. These results were supported when the proposed method was applied to both Arabic and English handwritten number databases.

In this paper the classification of motor imagery brain signals is addressed. The innovative idea is to use both temporal and spatial knowledge of the input data to increase the performance. Definitely, the electrode locations on the scalp is as important as the acquired temporal signals from every individual electrode. In order to incorporate this knowledge, a deep neural network is employed in this work. Both motor-imagery EEG and bi-modal EEG-fNIRS datasets were used for this purpose. The results are compared for different scenarios and using different methods. The achieved results are promising and imply that combining both temporal and spatial information of the brain signals could be really effective and increases the performance.

The problem of simultaneous blood oxygenation level dependent (BOLD) detection and data completion is addressed in this paper. It is assumed that a set of fMRI data with significant number of missing samples is available and the aim is to recover those samples with least possible quality degradation. At the same time, BOLD should be detected. We propose a new cost function comprising both BOLD detection and data reconstruction terms. A solution based on singular value thresholding and sparsity-inducing approach is proposed. Due to the low-rank nature of the fMRI data, it is expected that the related techniques to be very effective for reconstruction. Extensive experiments are conducted on different datasets in noisy conditions. The achieved results, both in terms of data quality and data analysis accuracy, are promising and confirm that the proposed method can be effective for recovery of compressed/incomplete fMRI data. Several state-of-the art image reconstruction techniques are compared with the proposed method. In addition, the results of applying general linear model (GLM) using statistical parameter mapping (SPM) toolbox are compared with those of the proposed method.