Yazar "Awad, Mohammed" seçeneğine göre listele
Listeleniyor 1 - 4 / 4
Sayfa Başına Sonuç
Sıralama seçenekleri
Öğe Classification of stroke using machine learning techniques : review study(IEEE, 2023) Sawan, Aktham; Awad, Mohammed; Qasrawi, RadwanAbstract—Presently, stroke is the leading cause of adult injury worldwide. The World Health Organization estimates that each year 15 million people around the world suffer a stroke. Five million of them die, and another five million are disabled for life. There is a chance to dramatically enhance the classification of strokes in the early stages. In this article, we reviewed all portable devices that produced electroencephalogram(EEG) data and all machine learning (ML) methods and deep-learning methods used to identify stroke using EEG data, and we noted that the amount of work on ML and deep learning in analyzing EEG data have increased rapidly in recent years. Such analysis has achieved greater precision compared to that conventional methods. We also discussed in this study the opportunities and key challenges for improving the accuracy of future work.Öğe Hybrid deep learning and metaheuristic model based stroke diagnosis system using electroencephalogram (EEG)(Elsevier, 2023) Sawan, Aktham; Awad, Mohammed; Qasrawi, Radwan; Sowan, MohammadOver the last few decades, there has been a significant increase in the average lifespan. Consequently, the number of elderly people suffering from strokes has also risen. As a result, strokes and their treatments have become crucial subjects of research, particularly for the application of machine learning. One of the primary factors in stroke treatment is the speed of response. Currently, both computed tomography (CT) and magnetic resonance imaging (MRI) are used to diagnose strokes. However, CT takes eight hours before an accurate diagnosis can be made, and MRI is expensive and not available in all hospitals. Therefore, there is a growing need for novel approaches to identifying strokes based on electroencephalogram (EEG) signals. In this paper, a hybrid model of deep learning and metaheuristic was developed in the offline stage to classify strokes. Since EEG data is a time series with frequencies, a hybrid model was deemed appropriate. This hybrid model combined a Convolutional Neural Network (CNN) with bidirectional Gated Recurrent Unit (BiGRU). The performance of this model surpassed that of other comparable models. Given the paramount importance of speed and accuracy in this work, the harmony search (HS) algorithm, which is specialized in handling frequencies, was used for feature selection. HS outperformed all similar algorithms when applied to the CNN-BiGRU hybrid model. Additionally, for the optimization of continuous hyperparameters, the multiverse optimization (MVO) algorithm was employed, which proved to be the most effective when compared to another similar algorithm for validation purposes. The new model, CNN-BiGRU-HS-MVO, was applied to analyze the data collected from Al Bashir Hospital using the MUSE-2 portable device, resulting in an impressive prediction accuracy of 99.991%. Moreover, it demonstrated an 11.08% improvement over the results from the paper titled “Predicting stroke severity with a 3-min recording from the Muse portable EEG study”. Furthermore, a decision support system was built on the cloud computing environment based on the hybrid model. This system allows for the diagnosis of patients anytime and from anywhere within minutes, with the authorized person receiving the diagnosis results through SMS notification.Öğe Machine learning-based approach for stroke classification using electroencephalogram (EEG) signals(Scitepress, 2022) Sawan, Aktham; Awad, Mohammed; Qasrawi, RadwanIn recent years, the health care field has heavily relied on the field of computation. The medical decision support system DSS, for instance, helps health professionals obtain accurate and reliable readings and diagnosis of patients’ vital signs. Nowadays, several medical devices allow capturing brain signals, some of these devices are wearable, which enhances signal quality and facilitates access to the signals than the traditional EEG devices. EEG signals are critical for assessing mental health and analyzing brain characteristics as they are able to detect a wide range of nerve-related diseases, such as stroke. This research seeks to study the use of machine learning techniques for the medical diagnosis of stroke through EEG signals obtained from the wearable device ‘MUSE 2.’ Eight ML techniques were used for analysis, the XGboost classifiers outperformed other classifiers in identifying strokes with an accuracy rate of 83.89%. The findings proved a 7.89% improvement on accuracy from the previous study “Predicting stroke severity with a 3-minute recording from the Muse portable EEG studyÖğe MACHINE LEARNING-BASED STROKE DISEASE DIAGNOSIS USING ELECTROENCEPHALOGRAM (EEG) SIGNALS(Taylors Univ Sdn Bhd, 2023) Sawan, Aktham f.; Awad, Mohammed; Qasrawi, Radwan; Sowan, MohammadStroke is currently ranked as the third leading cause of death worldwide. While computed tomography (CT) and magnetic resonance imaging (MRI) are commonly used for stroke diagnosis, they have their limitations. CT scans can be time-consuming, taking up to 8 hours to complete diagnosis, while MRI procedures can be lengthy, often making it impractical for most stroke patients. This has led to the necessity of exploring new methods for stroke detection, particularly utilizing EEG signals. In this paper, we propose a cloud computing -based machine learning (ML) system that leverages MUSE2 to diagnose stroke patients by analysing EEG signals. Our dataset, collected from Al Bashir Hospital between 2021 and 2022, consists of a randomly selected sample of 31 stroke patients and 31 healthy individuals. To pre-process the collected dataset, we employ Fourier and wavelet transformations. The processed EEG signals are then transmitted over the Internet to the ML model for stroke diagnosis. Real-time results are delivered to authorized personnel via SMS. During our research, various classifiers were evaluated, and a modified XGboost classifier emerged as the most effective choice. It outperformed other ML classifiers with an impressive accuracy of 96.87%.
