Dr. Alaa Taima

Software Defined Networks (SDNs) are one of the most important modern technologies in the field of networks, because of their advantages in the architecture and management of networks and control of their full functionality. SDN is distinguished from traditional networks by the presence of a central control element, which is the controller that is responsible for all operations that occur in the network. The controller is the main element that determines the success or failure of software-defined networks, so it was necessary to study and compare the different types of controllers that exist today. This paper proposes an empirical mathematical model to choose the best controller for SDN by using a Mininet emulator, concerning two performance metrics (Throughput and latency) for diverse parameters such as different types of topologies, diverse numbers of hosts, diverse numbers of switches, and diverse numbers of threads. These performance metrics have different weights depending on the needs of the users. We employ OpenFlow as a southbound protocol and five SDN controllers (Ryu, POX, OpenDaylight (ODL), and Floodlight). The results demonstrate that the suggested mathematical model is effective and flexible in choosing the best controller since the weights of performance measures are selected based on the needs of the user. The performance of the SDN network is better with ODL than with other SDN controllers.

Alzheimer's disease (AD) is a complex, progressive neurodegenerative condition that affects millions of people worldwide, making early diagnosis critical for effective treatment and clinical management to improve quality of life. In this study, we present an automated classification framework based on the Vision Transformer (ViT) model optimized with a modified hippopotamus optimization algorithm (M-HOA). Unlike traditional models that rely solely on ViTs or convolutional networks, the M-HOA algorithm is used to fine-tune key hyperparameters of the ViT model, improving feature extraction and classification accuracy. The model was evaluated on the ADNI dataset, which covers three diagnostic categories (AD, MCI, and NC). Experiments demonstrated that the proposed M-HOA-ViT model outperforms both the baseline and optimized ViT architectures, achieving a classification accuracy of 97.90%. The results indicate that integrating metaheuristic optimization with ViT significantly improves diagnostic accuracy, providing a robust and scalable approach for the early detection of Alzheimer's disease.

Abstract: The diagnosis of thyroid nodules remains a challenge due to the limitations of conventional imaging techniques. This paper aims to improve the accuracy and efficiency of thyroid nodule diagnosis. The proposed densnet121-IHOA model is a good solution to the diagnostic accuracy problem. The proposed model consists of a densely connected network to extract features from ultrasound images. Several layers are added to perform the diagnosis process based on the features extracted by Densnet121. The optimal hyper-parameters for learning rate, batch size, dropout ratio, and number of neurons were found using an optimization algorithm. The improved hippopotamus algorithm (IHOA) is efficient in finding hyper-parameters. The IHOA algorithm is robust in exploring and exploiting solutions to find optimal values, and it does not require a large number of iterations. The dataset used in this paper is AUITD. The number of images used in the paper was 2,121, divided into 1,697 training images and 424 test images. The proposed model achieved an accuracy of 97.7%, precision of 96.3%, recall of 98%, and F1 score of 97.4%.

Alzheimer's disease is a progressive neurodegenerative disorder that primarily affects individuals aged 65 and older, leading to irreversible memory loss and cognitive decline. Early detection plays a critical role in managing the disease and improving patient outcomes. In recent years, numerous studies have investigated the development of automated systems to identify the stages of Alzheimer's disease using advanced deep learning methods. This paper provides a structured literature review focused on the use of Vision Transformers (ViTs) and metaheuristic optimization algorithms for early diagnosis. The reviewed studies demonstrate that ViT-based models outperform traditional approaches in extracting spatial and temporal features from brain imaging data, achieving classification accuracies exceeding 96% on widely used datasets such as ADNI and OASIS. Additionally, the review addresses key challenges in processing 3D medical images and highlights ongoing efforts to develop hybrid architectures that integrate the strengths of Convolutional Neural Networks (CNNs) and Vision Transformers (ViT). The paper also explores how collaborative learning strategies can enhance model training while preserving patient privacy, making these techniques more suitable for real-world clinical applications.

This paper presents a survey of studies and research on machine learning techniques, deep learning, as well as research using optimization algorithms, and evolutionary algorithms, in relation to discoveries of diagnosing thyroid disorders. The paper includes an analysis of recent studies. The techniques of researchers in this field of diseases (thyroid diseases) are shown. As well as the results of the aforementioned studies on accuracy factors, Precision, Recall, F1-scor and. Advantages and disadvantages of advanced algorithms. By analyzing the current methods that were surveyed, it was proven that transfer learning techniques, as well as techniques that use optimization algorithms, are the most efficient. In some research, preprocessing plays a major role in obtaining better results in the model training stage.

Conventional feature selection methods frequently face difficulties in identifying the most significant features within high-dimensional data, which can result in lower classification accuracy and heightened computational demands. These methods typically do not consider the internal structure or harmony within the data, which limits their ability to identify optimal subsets of features. In this study, we address these limitations by proposing an efficient feature selection method, BER-WOA (Bayesian Expectile Regression-Whale Optimization Algorithm), for improving the performance of intrusion detection systems. The proposed approach consists of two stages: the first stage uses Bayesian Expectile Regression (BER) to select optimal features, reducing the search space. In the second stage, the Whale Optimization Algorithm (WOA) is applied to search within the smaller feature subset to identify the most relevant features for classification. The method is applied to two widely used datasets, NSL-KDD and UNSW-NB15, and evaluated using four popular classifiers: SVM, KNN, RF, and DT. The proposed BER-WOA method effectively reduces the number of features while maintaining high classification accuracy. On the NSL-KDD dataset, the method selected 18 features, achieving an accuracy of 99.14% with RF, while on the UNSW-NB15 dataset, 21 features were selected, yielding an accuracy of 99.72% with RF. BER-WOA method achieved even better results, with a 46.84% reduction in execution time for UNSW-NB15 and a 47.06% reduction for NSL-KDD. The results highlight that the BER-WOA method outperforms other f​

Death rates across the globe are often linked to respiratory illnesses, with severe conditions like­ chronic obstructive pulmonary disease (COPD) and asthma being the primary culprits. Early detection of the­se diseases in their initial stages is more crucial than we may realize. The ancient diagnostic technique of lung auscultation, where a stethoscope is placed on the lungs, is renowned but also has inherent limitations and susceptibility to data distortion due to environmental variables. This led to the deve­lopment of modern solutions, born out of necessity, to address these challenges innovative methods that harness the power of deep learning algorithms to capture respiratory sounds more accurately. The International Conference on Biomedical and Health Informatics (ICBHI) dataset, containing lung sound recordings, is available to the machine learning community for research and development. Leveraging machine learning and deep learning techniques, with the latter being a subset of machine learning, such as convolutional neural networks, has enabled more accurate diagnoses compared to traditional auscultation methods. These advanced algorithms have achieved impre­ssive voice classification accuracy rates, outperforming conventional approaches. The fusion of cutting-edge technology and medical expertise has the potential to revolutionize respiratory disease detection and management. Scientific investigations and research have demonstrated that when utilizing) ICBHI 2017 (data set, its precision varies from 42% to 90%. The goal of this article is to review articles related to the use of deep learning algorithms, which are combined in some articles with other machine learning algorithms, and the way they deal with the ICBHI 2017 dataset.

The use of lung sounds in conjunction with respiratory auscultation can aid in the diagnosis of abnormalities. There is the possibility for highly developed AI combined with deep learning to automate the study of sound. A technique that utilizes neural networks and genetic algorithms to classify lung sounds is provided here. The lung sounds of people suffering from a variety of pulmonary diseases as well as healthy subjects were recorded via the chest wall. A CNN-based method for categorizing respiration sounds is proposed here. This method makes use of current breakthroughs in the field of picture classification. The Mel Frequency Cepstral Coefficients (MFCCs) play a crucial role in converting audio signals into visual representations. The level of accuracy achieved while classifying respiratory sounds (Normal, Crackles, Wheezes, Both) is far higher than was anticipated. When it comes to determining lung sounds, ML and DL, and especially EA-optimized models, are superior to conventional methods such as chest CT scanning in terms of effectiveness.

Lung Cancer is one of the world’s peak-level types of cancer. There are so many people are died due to lung cancer diseases. Early identification of tumors would facilitate in sparing a vast number of lives globally consistently. Computed tomography (CT) is essential for pulmonary nodule detection in diagnosing lung cancer. Recently, for creating medical image classification problems, machine learning (ML) and deep learning (DL) have emerged as two of the most popular and commonly utilized approaches. Furthermore, employing chest CT scans from pneumonia patients, ML and DL models outperformed conventional approaches in effectiveness. The structure and parameters of ML and DL models can be optimized using evolutionary algorithms (EA). This survey aims to overview the recent works in lung cancer detection-based ML and DL models that EA has optimized.

One of the factors that kills hundreds of people every year is driving accidents caused by drowsy drivers. There are different methods to prevent this type of accidents. Recently Machine Learning (ML) and Deep Learning (DL) have emerged as very effective and valuable approaches for detecting driver drowsiness. Moreover, the optimization of machine learning (ML) and deep learning (DL) models may be achieved through the utilization of evolutionary algorithms (EA). This survey aims to offer an overview of recent studies in driver drowsiness detection-based machine learning and deep learning models that have been improved by EA. This survey divides the approaches for detecting drowsiness into two groups: those that rely on ML, and DL, and those that rely on models-based deep learning and machine learning that are optimized by evolutionary algorithms.

Driver drowsiness is among the main causes of fatal traffic accidents or injuries of varying severity around the world. Every year, numerous studies and research projects address this topic. Considering the enormous potential of artificial intelligence especially the deep learning technologies, the current focus is on developing deep convolutional neural networks (CNNs) specifically designed to detect driver drowsiness. Designing these networks is difficult because most methods rely on experimentation and optimization to determine hyperparameter values. This article presents a method to enhance the hyperparameters of a convolutional neural network by using a modified version of the Walrus Optimization Algorithm (WaOA). The algorithm (M-WaOA) incorporates a logistic map to prepare the primary generation and the Mantegna’s algorithm accelerating access to the optimal solution. to Improve drowsiness detection system accuracy. The YAWDD dataset is used to obtain driving-related videos to achieve this goal. The footage is converted into individual frames, and a facial recognition algorithm is used to recognize the driver’s face. Furthermore, employing CNN to classify individuals as alert, sleepy, or asleep. The proposed technique achieved an accuracy of up to

Using 2D scans or simple 3D convolutions are two limitations of previous works on segmentation of brain tumors by deep learning, which lead to ignoring the temporal distribution of the scans. This study proposes a novel extension to the well-known U-net model for brain tumor segmentation, utilizing 3D Magnetic Resonance Imaging (MRI) volumes as inputs. The method, called ConvLSTM-based U-net + up skip connections, incorporates the ConvLSTM blocks to capture spatio-temporal dependencies in the 3D MRI volumes, and up skip connections to capture low-level feature maps extracted from the encoding path, enhancing the information flow through the network to the standard U-net architecture. A novel intensity normalization technique is used to improve the comparability of scans. This technique normalizes image intensity by subtracting the grey-value of the most frequent bin from the image. The novel method is tested on the Multimodal Brain Tumor Segmentation (BRATS) 2015 dataset, showing that the use of ConvLSTM blocks improved segmentation quality by 1.6% on the test subset. The addition of skip connections further improved performance by 3.3% and 1.7% relative to the U-net and ConvLSTM-based U-net models, respectively. Moreover, the inclusion of up skip connections could enhance the performance by 5.7%, 3.99% and 2.2% relative to the simple U-net, ConvLSTM-based U-net, and ConvLSTM-based U-net with skip connections, respectively. Finally, the novel preprocessing technique had a positive effect on the proposed network, resulting in a 3.3% increase in the segmentation outcomes.

Software-defined network (SDN) is based on separating data plane from control plane and decision-making processes to centralize network control. The OpenFlow (OF) protocol is the most common and widely used in software-defined network for communicating and controlling switches. With this protocol, the switch learns the routing information from the controller and then passes data packets based on this information. One of the most important components of the SDN is the controller, which is the smartest component of the network. Many controllers have been developed since the advent of SDN. One of the most important and famous controllers is Ryu. Due to the importance of the Ryu controller in SDN, in this article, the performance of the Ryu controller in terms of latency and throughput are evaluated. Cbench is used for measuring throughput and latency of this controller.

Software Defined Networks (SDNs) are one of the most important modern technologies in the field of networks, because of their advantages in the architecture and management of networks and control of their full functionality. SDN is distinguished from traditional networks by the presence of a central control element, which is the controller that is responsible for all operations that occur in the network. The controller is the main element that determines the success or failure of software-defined networks, so it was necessary to study and compare the different types of controllers that exist today. This paper proposes an empirical mathematical model to choose the best controller for SDN by using a Mininet emulator, concerning two performance metrics (Throughput and latency) for diverse parameters such as different types of topologies, diverse numbers of hosts, diverse numbers of switches, and diverse numbers of threads. These performance metrics have different weights depending on the needs of the users. We employ OpenFlow as a southbound protocol and five SDN controllers (Ryu, POX, OpenDaylight (ODL), and Floodlight). The results demonstrate that the suggested mathematical model is effective and flexible in choosing the best controller since the weights of performance measures are selected based on the needs of the user. The performance of the SDN network is better with ODL than with other SDN controllers.