الفهرس 
chapter one : Introduction and Literature SurveyOnly 14 pages are availabe for public view
 |  | from | 194 |  |  |
| | | from | 194 | | |
Abstract
Multiple Sclerosis (MS) is a chronic autoimmune disorder that affects the central nervous system, leading to progressive neurological impairment. This disease, which has a multifactorial etiology including genetic and environmental factors, affects around 2.8 million people worldwide and is more prevalent in females. The clinical manifestations of MS are diverse, ranging from visual disturbances and muscle weakness to cognitive impairments and emotional changes, which can significantly impact patients’ quality of life. MS is primarily diagnosed through Magnetic Resonance Imaging (MRI). However, the process of interpreting MRI data is time-consuming, prone to errors, and requires significant expertise, especially when differentiating MS from other similar demyelinating diseases. This underscores the need for innovative diagnostic approaches to enhance accuracy and reduce diagnosis time.This study proposes two advanced AI-based methodologies for the automatic identification of MS lesions from MRI scans. The first approach introduces an AI technique that uses feature extraction from brain MRI images combined with an innovative feature selection method to achieve a high detection accuracy of 92.94%. The second approach presents a hybrid AI model that integrates a multi-view ResNet architecture and novel attention mechanisms for feature extraction, along with a new metaheuristic algorithm for dimensionality reduction, further enhancing diagnostic precision and efficiency, achieving an impressive accuracy of 98.29%.The thesis is organized into several chapters, each focusing on different aspects of the research. Chapter 1 Provides a concise overview of MS, including its definition, historical background, and prevalence. It explains the different types of MS and the disease’s pathophysiology, covering immune responses, demyelination, and lesion formation. The chapter discusses common symptoms, diagnostic criteria, and the importance of MRI and laboratory tests. It also outlines current treatments, such as disease-modifying therapies, and highlights challenges in diagnosis and management. The relevant literature survey has been explored to indicate the importance of our study and to show the progress of AI techniques for diagnosing MS.Chapter 2 explores key medical imaging techniques in healthcare. It introduces the medical imaging modalities and their significance. The chapter covers CT concepts, image acquisition, and their pros and cons. A thorough discussion on MRI follows, outlining its principles, hardware components, image acquisition techniques, and safety considerations. The chapter concludes with the basic physics of medical imaging modalities, focusing on MRI as it’s the main imaging technique used in the diagnosis of MS.In the third chapter, we provide the fundamentals of image processing with machine and deep learning techniques. It begins with the various phases of medical image preparation. The definitions of feature extraction, feature selection, and classifiers are then presented to illustrate the different methods of classification based on machine learning and deep learning.Chapter 4 introduces an innovative MRI-based method for diagnosing MS. This method extracts features from brain MRI images using first-order histograms, the gray level co-occurrence matrix, and local binary patterns, then selects features with the Sine Cosine Algorithm and Sea-horse Optimizer. Tested on the eHealth lab dataset, it achieved 97.97% accuracy with the k-nearest neighbors classifier. Validation on a larger dataset showed 92.94% accuracy for FLAIR images and 91.25% for T2-weighted images using Random Forest, outperforming existing MS detection methods.Finally, chapter 5 presents an advanced hybrid AI model to improve MS diagnosis accuracy using MRI scans. It combines a multi-view ResNet architecture with attention mechanisms (VSAB and VCAB) for extracting features from 2D brain images. The Quantum RIME strategy, merging RIME and QPSO, enhances classification accuracy and efficiency. Tested on sixteen UCI benchmark datasets and 425 brain MRI scans, the model achieved 98.29% accuracy, 96.49% precision, 97.65% specificity, and 97.85% F1-score, showcasing its potential in medical diagnostics for neurological disorders.