Facial Expression Recognition (FER) is a pivotal task in computer vision with applications spanning human-computer interaction, mental health monitoring, and security. While Convolutional Neural Networks (CNNs) have long dominated this field, the emergence of Transformer architectures has introduced new paradigms for capturing global dependencies. This paper presents a comparative study between two prominent transformer algorithms: the Vision Transformer (ViT) and the Swin Transformer, evaluated on the enhanced Real-world Affective Faces Database (RAF-DB). Our experimental results demonstrate that both architectures achieve exceptional recognition performance, with test accuracies exceeding 97%. Specifically, ViT achieved a test accuracy of $97.79 \%$, marginally outperforming Swin Transformer’s $97.70 \%$. This study highlights the robustness of Transformer-based models in capturing discriminative facial features when combined with effective preprocessing pipelines.

Malaria is a life-threatening infectious disease caused by Plasmodium parasites and can become severe if not diagnosed and treated promptly. Conventional diagnostic methods are often time-consuming, costly, and susceptible to human error. Computer-aided diagnostic systems provide a faster and more objective alternative for detecting malaria parasites from microscopic images. With the growing availability of medical imaging data, artificial intelligence–based approaches have become increasingly suitable for automated malaria classification. This study proposes a structured two-level fusion framework for binary malaria cell classification. At the first level, source fusion integrates original, Gaussian-filtered, and median-filtered images to enhance feature diversity. At the second level, decision fusion combines probabilistic outputs from an ensemble of Vision Transformer and Convolutional Neural Network architectures using sum rule, product rule, and majority voting strategies. The classification stage employs Vision Transformer Tiny, GhostNet100, and Mobile Vision Transformer Extra Small models. The contribution of this work lies in the systematic hierarchical integration of source-level diversity and structured probabilistic decision aggregation across heterogeneous lightweight architectures. Experimental results demonstrate that the proposed approach improves classification robustness and performance. The product rule within the two-level fusion framework achieves superior performance, reaching 98.9

Facial Expression Recognition (FER) is a fundamental problem in affective computing and human– computer interaction, with applications in healthcare, surveillance, and intelligent user interfaces. Recent advances in deep learning, particularly convolutional neural networks (CNNs), have significantly improved FER performance by leveraging transfer learning from large-scale image datasets. This paper presents a comparative performance analysis of four state-of-the-art pre-trained CNN architectures, namely, ResNet152, EfficientNetV2-L, Xception, and DenseNet201 which are applied to facial expression recognition using the Extended Cohn–Kanade (CK+) dataset. The experiments were conducted on eight emotion classes with data augmentation applied to enhance generalization. All models were fine-tuned using the Adam optimizer under identical experimental conditions. Performance evaluation was carried out using confusion matrices and class-wise, micro-averaged, and weighted-averaged metrics. Among the evaluated models, DenseNet201 achieved the highest accuracy of 89.25%, outperforming EffiecientNetV2-L by 1.08% under identical training conditions. This indicates that dense feature reuse is particularly effective for capturing subtle facial expression patterns, while EfficientNetV2-L provides a balanced trade-off between accuracy and model efficiency

The research in this paper leverages Electroencephalography (EEG) to develop a Brain-Computer Interface (BCI). By processing and analyzing EEG signals, the system translates brain activity into commands, enabling individuals with severe motor impairments to control devices and regain independence. The Berlin BCI Competition IV dataset is used in the study. EEG recordings of healthy subjects doing hand and foot motor imagery tasks are included in the collection. Discrete wavelet transformation (DWT) decomposes the EEG signals into sub-bands to obtain pertinent information. Feature selection is performed using the ant colony optimization algorithm (ACOA) to reduce dimensionality and improve classification accuracy. An evaluation of three robust classifiers-support vector machine (SVM), weighted k-nearest neighbors (Weighted k-NN), and wide neural networks (WNN)-is conducted to ascertain their capability in Motor Imagery (MI) task classification. The results demonstrate that the SVM classifier in conjunction with DWT and ACO, achieves the highest average accuracy score of 85.36% while achieving a dimension reduction of 2.064-times. It makes it an appealing approach for the BCI systems.

In recent advancements in cancer diagnosis from MRI imaging, pre-trained deep learning models have shown promise for improving accuracy, efficiency, and scalability. By employing transfer learning, these DL models can leverage their existing knowledge from extensive image-based datasets to adapt to the specific patterns found in MRI scans. In this paper, we compare the performance of MobileNet, ResNet, DenseNet, and EfficientNet models, highlighting the varying strengths of these architectures. We also propose two different ensemble models to improve the accuracy and efficiency of the classification. Extensive experiments on the Br35H brain MRI database demonstrated that these ensemble models exhibit transfer learning capabilities and optimize diagnostic performance. These insights highlight the value of tailored pre-trained models and ensemble methods in enhancing MRI-based cancer diagnostics.

The work in this paper presents an advanced method for classifying electrocardiogram (ECG) signals into four cardiac disease categories using feature extraction transfer learning with convolutional neural networks (CNNs). Addressing the need for accurate automated ECG diagnosis, the study develops a custom 5-layer CNN model. A key improvement is the feature extraction transfer learning strategy, where the convolutional layers served as feature extractors and the extracted features were subsequently classified using either the CNN’s fully connected layers, a Support Vector Machine (SVM) classifier, or a Random Forest (RF) classifier. This hybrid approach leverages CNNs for feature extraction and SVM/RF for final classification. Experimental results show the proposed method’s superiority in classification accuracy, training efficiency, and generalization ability. These improvements highlight the framework’s potential to enhance automated ECG interpretation and improve patient outcomes significantly.

Automated and accurate brain tumor classification from MRI scans is a promising application of deep learning. This paper presents a YOLOv11-based deep learning model for detecting and classifying three tumor types: glioma, meningioma, and pituitary. Performance evaluation demonstrates the model's strong capability in tumor detection and classification, particularly with meningioma and pituitary tumors showing higher precision than glioma. Validation curves indicate steady reduction in loss functions across epochs, signifying effective learning and convergence. The model was trained and validated on a structured dataset, achieving a high accuracy performance of 98.9%. Deep learning-based tumor classification can facilitate early detection, assist radiologists, and improve clinical decision-making process.

Malaria is an infectious disease caused by Plasmodium parasites, and it can be severe if left untreated. Traditional diagnostic methods are costly and prone to human error. However, computer-aided techniques offer a faster and more accurate way to detect malaria viruses. With the vast amounts of available data, deep learning approaches are particularly suited for malaria classification. This paper introduces a method combining original and filtered data through source fusion, improving classification accuracy by up to 7%. Additionally, it proposes fusing the probabilistic decisions of an ensemble of pre-trained CNN classifiers using both original and filtered data. Extensive experiments conducted on a well-known malaria dataset demonstrate that the proposed approach achieves exceptional classification performance, with 100% accuracy, specificity, and sensitivity.

This paper presents the design and development of a dry-EEG-based motor imagery classification system for a Brain-Computer Interface (BCI)-operated wheelchair, utilizing a deep learning approach. While wet EEG remains prevalent for brain activity recording, its discomfort limits continuous monitoring. Dry EEG offers improved comfort and ease of use, prompting this study to explore its potential. We propose a system that classifies dry-EEG signals into four motor imagery classes. Our objective is to extract meaningful features from the inherently noisy and weak dry-EEG data, thereby enhancing performance for practical BCI applications. A Convolutional Neural Network (CNN) model was developed, and the impact of Principal Component Analysis (PCA) for dimensionality reduction and feature extraction was evaluated. Experimental results demonstrated classification accuracies of 93.33% using raw data and 96.42% after preprocessing the data by PCA, showing the system's effectiveness.

Ethnicity recognition using artificial intelligence has emerged as a significant application in computer vision. It relies on deep learning techniques to classify individuals based on facial features. In this research, we propose a CNN-based model to classify a person’s ethnicity using a face dataset with diverse age, gender, and ethnicity data. The dataset contains 27,305 face images organized into 5 different ethnicities. A Haar cascade classifier has been implemented in the system for face detection, allowing for quick extraction of the region of interest before passing the images into the CNN model. To improve performance, checkpoints were used to save the besttrained model. An accuracy of $96.11 \%$ was achieved by the CNN model for ethnicity recognition. This proves the expediency and practicality of the proposed deep learningbased ethnicity and gender recognition models.

In this paper, we studied and compared the benefits of applying transformation as a preprocessing step to the input of the deep learning algorithms. A comparison was conducted between the discrete wavelet transform (DWT) and fast Fourier transform (FFT) as feature extractors for the ECG signals. The generated feature vectors were fed to a convolutional neural network and long short-term memory (CNN-LSTM) model for classification. Extensive experiments were executed on ECG signals collected from MIT-BIH and BIDMC databases with 3 classes. The obtained results showed the combined power of transformation and deep learning which led to better classification accuracies.

This paper introduces a novel deep learning model for ECG signal classification using feature fusion. The proposed methodology transforms the ECG time series into a spectrogram image using a short-time Fourier transform (STFT). This spectrogram is further processed to generate a histogram of oriented gradients (HOG) and local binary pattern (LBP) features. Three separate 2D convolutional neural networks (CNNs) then analyze these three image representations in parallel. To enhance performance, the extracted features are concatenated before feeding them into a gated recurrent unit (GRU) model. The proposed approach is extensively evaluated on two ECG datasets (MIT-BIH + BIDMC and MIT-BIH) with three and five classes, respectively. The experimental results demonstrate that the proposed approach achieves superior classification accuracy compared to existing algorithms in the literature. This suggests that the model has the potential to be a valuable tool for accurate ECG signal classification, aiding in the diagnosis and treatment of various cardiovascular disorders.

Breast cancer, which has the highest mortality and morbidity rates among diseases affecting women, poses a significant threat to their lives and health. Early diagnosis is crucial for effective treatment. Recent advancements in artificial intelligence have enabled innovative techniques for early breast cancer detection. Convolutional neural networks (CNNs) and support vector machines (SVMs) have been used in computer-aided diagnosis (CAD) systems to identify breast tumors from mammograms. However, existing methods often face challenges in accuracy and reliability across diverse diagnostic scenarios. This paper proposes a three parallel channel artificial intelligence-based system. First, SVM distinguishes between different tumor types using local binary pattern (LBP) features. Second, a pre-trained CNN extracts features, and SVM identifies potential tumors. Third, a newly developed CNN is trained and used to classify mammogram images. Finally, a decision fusion that combines results from the three channels to enhance system performance is implemented using different rules. The proposed decision fusion-based system outperforms state-of-the-art alternatives with an overall accuracy of 99.1% using the product rule.

Cardiovascular diseases (CVDs) like arrhythmia and heart failure remain the world’s leading cause of death. These conditions can be triggered by high blood pressure, diabetes, and simply the passage of time. The early detection of these heart issues, despite substantial advancements in artificial intelligence (AI) and technology, is still a significant challenge. This research addresses this hurdle by developing a deep-learning-based system that is capable of predicting arrhythmias and heart failure from abnormalities in electrocardiogram (ECG) signals. The system leverages a model that combines long short-term memory (LSTM) networks with convolutional neural networks (CNNs). Extensive experiments were conducted using ECG data from both the MIT-BIH and BIDMC databases under two scenarios. The first scenario employed data from five distinct ECG classes, while the second focused on classifying data from three classes. The results from both scenarios demonstrated that the proposed deep-learning-based classification approach outperformed existing methods.

Heart disease, a global killer with many variations like arrhythmia and heart failure, remains a major health concern. Traditional risk factors include age, cholesterol, diabetes, and blood pressure. Fortunately, artificial intelligence (AI) offers a promising solution. We have harnessed the power of AI, specifically deep learning and convolutional neural networks (CNNs), to develop Rhythmi, an innovative mobile ECG diagnosis device for heart disease detection. Rhythmi leverages extensive medical data from databases like MIT-BIH and BIDMC. These data empower the training and testing of the developed deep learning model to analyze ECG signals with accuracy, precision, sensitivity, specificity, and F1-score in identifying arrhythmias and other heart conditions, with performances reaching 98.52%, 98.55%, 98.52%, 99.26%, and 98.52%, respectively. Moreover, we tested Rhythmi in real time using a mobile device with a single-lead ECG sensor. This user-friendly prototype captures the ECG signal, transmits it to Rhythmi’s dedicated website, and provides instant diagnosis and feedback on the patient’s heart health. The developed mobile ECG diagnosis device addresses the main problems of traditional ECG diagnostic devices such as accessibility, cost, mobility, complexity, and data integration. However, we believe that despite the promising results, our system will still need intensive clinical validation in the future.

The use of data fusion can be of a enormous help in boosting classification performance. Feature fusion is a data fusion technique that is being considered in this study. The effect of fusing different feature descriptors extracted by using histogram-based local feature extraction algorithms on the performance of the face recognition problem is investigated. Feature fusion/concatenation of more than one generated feature descriptor is applied. The impact of fused two and three feature descriptors on the system performance is evaluated when the training set is limited to only one-shot per person. Extensive experiments are carried out using two well-known face databases. Comparisons are conducted among different algorithms for extraction of the local statistical feature descriptors of the face images. The obtained results show that feature fusion of the descriptors can significantly improve the performance with certain feature descriptors.

The threat of heart disease has increased throughout history due to the difficulty of diagnosing it and the need for experienced doctors. Heart disease can be caused by various of factors such as age, high blood pressure, and diabetes. This paper investigates the arrythmia of Electrocardiogram (ECG) by comparing and evaluating different classifiers to predict the abnormalities of a heartbeat. We have developed a system that predicts heartbeat arrhythmia using a multi-classifier deep learning model. By using MIT-BIH Arrythmia Database, we were able to achieve an accuracy of 97.4%. Compared to other models that we have implemented; this model recorded the highest performance.

This study will detail the effects of computer-based graphic design on illustration in poster works. The understanding of art has evolved from ancient times to the present, giving rise to design works. With the development of the internet and computer technologies in the 20th century, the understanding of design has also changed. The impact of computer technology on the drawings in poster works will be highlighted in this study as how they began in the 1950s and 1960s and developed throughout the century. Especially in the 1990s, technological developments offered new technical possibilities to art and artists. The production of computer-aided unique designs and illustrations began. The study's evaluation portion will cover the technical aspects of the illustrations seen in the posters created by well-known historical designers. The examples presented throughout the study cover the effects of computer-based graphic design in the 20th century on the illustration in poster works.

The use of local statistical descriptors for image representation has emerged and gained a reputation as a powerful approach in the last couple of decades. Many algorithms have been proposed and applied, since then, in various application areas employing different datasets, classifiers, and testing parameters. In this paper, we felt the need to make a comprehensive study of frequently-used statistical local descriptors. We investigate the effect of using different histogram-based local feature extraction algorithms on the performance of the face recognition problem. Comparisons are conducted among 18 different algorithms. These algorithms are used for the extraction of the local statistical feature descriptors of the face images. Moreover, feature fusion/concatenation of different combinations of generated feature descriptors is applied, and the relevant impact on the system performance is evaluated. Comprehensive experiments are carried out using two well-known face databases with identical experimental settings. The obtained results indicate that the fusion of the descriptors can significantly enhance the system's performance.

The year 2020 will certainly be remembered in human history as the year in which humans faced a global pandemic that drastically affected every living soul on planet earth. The COVID-19 pandemic certainly had a massive impact on human's social and daily lives. The economy and relations of all countries were also radically impacted. Due to such unexpected situations, healthcare systems either collapsed or failed under colossal pressure to cope with the overwhelming numbers of patients arriving at emergency rooms and intensive care units. The COVID -19 tests used for diagnosis were expensive, slow, and gave indecisive results. Unfortunately, such a hindered diagnosis of the infection prevented abrupt isolation of the infected people which, in turn, caused the rapid spread of the virus. In this paper, we proposed the use of cost-effective X-ray images in diagnosing COVID-19 patients. Compared to other imaging modalities, X-ray imaging is available in most healthcare units. Deep learning was used for feature extraction and classification by implementing a multi-stream convolutional neural network model. The model extracts and concatenates features from its three inputs, namely; grayscale, local binary patterns, and histograms of oriented gradients images. Extensive experiments using fivefold cross-validation were carried out on a publicly available X-ray database with 3886 images of three classes. Obtained results outperform the results of other algorithms with an accuracy of 97.76%. The results also show that the proposed model can make a significant contribution to the rapidly increasing workload in health systems with an artificial intelligence-based automatic diagnosis tool.