Saudi Cultural Missions Theses & Dissertations
Permanent URI for this communityhttps://drepo.sdl.edu.sa/handle/20.500.14154/10
Browse
14 results
Search Results
Item Restricted SEVERITY GRADING AND EARLY DETECTION OF ALZHEIMER’S DISEASE THROUGH TRANSFER LEARNING(Saudi Digital Library, 2025) Alqahtani, Saeed; Zohdy, MohamedAlzheimer’s disease (AD) is a neurological disorder that predominantly affects individuals aged 65 and older. It is one of the primary causes of dementia, and it contributes significantly and progressively to impairing and destroying brain cells. Recently, efforts to mitigate the impact of AD have focused with particular emphasis on early detection through computer aided diagnosis (CAD) tools. This study aims to develop deep learning models for the early detection and classification of AD cases into four categories: non-demented, moderate-demented, mild-demented, and very mild demented. Using Transfer Learning technique (TL), several models were implemented including AlexNet, ResNet-50, GoogleNet (InceptionV3), and SqueezeNet, by leveraging magnetic resonance images (MRI) and applying image augmentation techniques. A total of 12,800 images across the four classifications that were preprocessed to ensure balance and meet the specific requirements of each model. The dataset was split into 80% for training and 20% for testing. AlexNet achieved an average accuracy of 98.05%, GoogleNet (InceptionV3) reached 97.80%, ResNet-50 attained 91.11%, and SqueezeNet 86.37%. The use of transfer learning method addresses data limitations, allowing effective model training without the need for building from scratch, thereby enhancing the potential for early and accurate diagnosis of Alzheimer’s disease [1].12 0Item Restricted Improving Sleep Health with Deep Learning: Automated Classification of Sleep Stages and Detection of Sleep Disorders(Saudi Digital Library, 2024-07-07) Almutairi, Haifa; Datta, AmitavaSleep consumes roughly one-third of a person’s lifetime, and it is characterized by distinct stages within sleep cycle. The sequence of these stages at night provides insights into the quality of sleep. Poor sleep quality can have numerous consequences, including drowsiness, reduced concentration, and fatigue. Beyond sleep quality, an analysis of the sequence of sleep stages can uncover the presence of sleep disorders. This thesis aims to focus on three key research problems related to sleep. Firstly, it focuses on the classification of sleep stages using a combination of signals and deep learning models. Sleep stages are categorized into five distinct stages, namely Wake (W), non-rapid eye movement (NREM) stages comprising N1, N2, and N3, and rapid eye movement (REM) stage. Throughout the duration of sleep, individuals experience multiple cycles of sleep stages. Each cycle contains a standard allocation of each stage. An unbalanced distribution of the stages can indicate the presence of sleep disorders. Previous studies primarily classified sleep stages using a single channel of electroencephalography (EEG) signals. However, incorporating a combination of signals from electromyography (EMG) and electrooculogram (EOG) alongside EEG data provides additional features. These features extracted from muscle activity and eye movements during sleep, thereby enhancing classification accuracy. In this thesis, a robust model called SSNet is proposed to accurately classify sleep stages from a fusion of EEG, EMG, and EOG signals. This model combine convolutional neural networks (CNNs) and long short-term memory (LSTM) networks to extract the salient features from various physiological signals. The CNN architecture extracts spatial features from the input signals, while LSTM architecture captures the temporal features present in signals. This study has obtained encouraging outcomes in the classification of sleep stages through the fusion of physiological signals and deep learning techniques. Secondly, this thesis aim to detect obstructive sleep apnoea (OSA) from electrocardiography (ECG) signals using deep learning methods. Sleep disorder breathing (SDB) is categorized into three different types, which are OSA, central sleep apnoea, and mixed sleep apnoea. OSA is the most common form of SDB that is characterized by repeated interruptions in breathing during sleep, leading to fragmented sleep patterns and various health complications. Previous studies developed feature engineering methods and machine learning models for the detection of OSA. Feature engineering methods involve crafting relevant features to feed into machine learning models. However, feature engineering is time-consuming and requires domain expertise. In contrast, deep learning automatically extracts features from ECG signals for OSA detection, eliminating the need for manual feature engineering methods. In this thesis, three deep learning architectures are proposed, including standalone convolutional neural networks (CNN), CNN with long short-term memory (LSTM), and CNN with gated recurrent unit (GRU). Through rigorous experimentation and evaluation, the combination of CNN and LSTM architecture is the best-performing model for OSA detection. To further enhance the architecture’s performance, the hyperparameters of the CNN and LSTM models were tuned and tested over a large dataset to validate their effectiveness. The third research problem addressed in this thesis is detection of periodic leg movements (PLM) and SDB from NREM stage by using a combination of signals and deep learning models. PLM is characterized by involuntary leg movements during sleep. These movements can disrupt sleep and result in daytime sleepiness with reduced quality of life. Detecting PLM and SDB events during NREM stage allows for quantifying the severity of sleep disorders. Previous studies have focused on the development of signal-based models for detecting PLM or SDB. However, the models lacked the ability to distinguish these events within specific sleep stages. To address this problem, a novel deep learning architecture known as DeepSDBPLM is proposed. This architecture aims to detect PLM and SDB events during the NREM stage. This architecture incorporates novel input features called attention EMDRaw signals and utilizes a Residual Convolutional Neural Network (ResCNN) model. This thesis presents experimental results using publicly available datasets to evaluate the performance of the proposed deep learning models for classification of sleep stages, and detection of sleep disorders. The models were evaluated standard metrics. It includes accuracy, sensitivity, specificity, and F1 score. The empirical results establish the effectiveness of proposed approaches. The models can be a stepping stone towards more advanced techniques.11 0Item Restricted Features Selection Strategies for Classifying Heterogeneous Cardiovascular Disease Data(University of Liverpool, 2025-05) Aldosari, Hanadi; Conene, Frans; Zheng, YalinThe exponential growth of data across diverse domains has presented significant challenges, particularly in the integration and analysis of multi-modal datasets. This thesis addresses these challenges by proposing a Homogeneous Feature Vector Representation (HFVR) framework, designed to unify disparate data formats and facilitate holistic machine learning. The primary application of this research is the classification of Cardiovascular Disease (CVD), using data from multiple sources, including time series, images, video, and clinical records. The main research question guiding this thesis is: How can multiple heterogeneous data sources be effectively combined to support comprehensive and integrated machine learning analysis? The contributions of this work are twofold: (i) the development of technical methodologies for feature extraction, and (ii) the application of these methods within a medical context to enhance CVD diagnosis and prognosis. Five feature extraction techniques are presented to address the complexities of multi-modal data integration: 1-1D Motifs and Discords (1D-MD): This technique uses matrix profiles to extract recurring patterns (motifs) and anomalies (discords) from ECG time series data. It serves as a benchmark for classification models. 2-2D Motifs and Discords (2D-MD): This technique operates directly on ECG images to extract spatial motifs and discords, offering improvements in classification performance over 1D-MD. 3-2D Convolutional Neural Networks (2D-CNN): Pre-trained CNNs like ResNet-50 and VGG16 are used to extract hierarchical features from ECG images, significantly improving classification accuracy when integrated into the HFVR framework. 4-Multi-Frame 2D CNN (MF2D-CNN): Designed for video data, this method processes frames using CNNs and applies temporal aggregation to capture dynamic patterns while maintaining computational efficiency. 5-Spatio-Temporal 3D CNN (ST3D-CNN): Building on MF2D-CNN, this approach uses 3D convolutions to jointly analyze spatial and temporal dynamics in Echo video data. The research also includes the development of a bespoke, multi-modal dataset in collaboration with the Liverpool Heart and Chest Hospital (LHCH), combining ECG, Echo, and clinical data. This dataset was used to evaluate the proposed methods and demonstrate their real-world applicability. The HFVR framework outperformed single-modality approaches by integrating features from multiple data sources. Extensive experiments were conducted on public datasets (CPSC, GHS, GAF, eCAN, and dCAN) and the LHCH dataset. Evaluation metrics such as Accuracy, Precision, Recall, F1-score, and AUC were used, with ten-fold cross-validation and stratified sampling ensuring robustness. Traditional classifiers like Support Vector Machines (SVMs) and k-Nearest Neighbour (kNN) were also used to validate the HFVR framework. Results showed that combining MF2D-CNN with clinical and 2D-CNN features achieved the highest AUC of 93.3%, significantly outperforming baseline methods. Statistical analysis confirmed the robustness and scalability of the techniques. Overall, this thesis advances multi-modal data integration by presenting a unified framework for feature extraction and fusion. The HFVR framework paves the way for holistic machine learning and improved predictive accuracy. While focused on CVD classification, the techniques are generalizable to other domains. Future work will explore real-time implementation, enhanced extraction methods, and expansion to additional data types and domains, representing a major step toward scalable, integrated machine learning systems.12 0Item Restricted Enhancing Lip Synchronization in Deep Learning Models: An Evaluation of Supplementary Metrics for Wav2Lip Performance Optimization(Queen Mary University of London, 2025) Almelabi, Mohammed; Naich, Ammar YasirThe technology of lip synchronization aims at lip movements in videos with corresponding audio and has proven itself to be extremely useful in multimedia applications. The Wav2Lip model leverages deep learning to achieve high-quality lip-syncing videos that have become a leading approach in this field. This paper investigates the use of different evaluation metrics in assessing the performance of the Wav2Lip model. The purpose of this analysis is to improve the loss metric in training the loss function in training the model and provide insights into improving the development of lip synchronization models for more realistic results.14 0Item Restricted Aspect-Based Sentiment Analysis on Healthcare Services Uding pre-trained Languges Model(Malaya University, 2025) Alkathiri, Sarah; Sabri, AznulThis research explores the application of various computational models for aspect- based sentiment analysis (ABSA) of healthcare reviews, a critical component of enhancing healthcare services through feedback analysis. With the rapid expansion of online health platforms, the volume of textual reviews generated by patients provides a rich source of data for understanding patient satisfaction and areas needing improvement. The research thoroughly assesses various models, encompassing conventional statistical models, recurrent neural networks (RNNs), and sophisticated transformer-based models like BERT, RoBERTa, and DistilBERT. Each model was assessed based on its ability to accurately classify sentiments tied to specific aspects of healthcare services, such as cleanliness, staff behavior, and treatment efficacy. Two primary feature extraction techniques, Bag of Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF), were employed to transform raw text into a suitable format for model ingestion. Our findings demonstrate that while traditional models offer quick and interpretable results, they sometimes lack the nuanced understanding of context provided by more sophisticated deep learning and transformer models. RNNs, particularly LSTM and BiLSTM, were effective in capturing temporal dependencies in text data, essential for comprehending longer patient feedback.29 0Item Restricted Intelligent Diabetes Screening with Advanced Analytics(University of Birmingham, 2024) Aldossary, Soha; Smith, PhillipDiabetes mellitus is a prevalent chronic disease with significant health implications worldwide. This project aimed to mitigate this pressing public health concern by using machine learning techniques and deep learning algorithms. I also established an online platform at which patients can enter their test results and health information and receive real-time diabetes detection and dietary recommendations based on their health profiles. Research has illustrated that models such as Gradient Boosting, Random Forest and Decision Trees perform well in diabetes prediction due to their ability to capture complex nonlinear relationships and handle diverse input features. Therefore, this project incorporated these models with others, such as the Support Vector Classifier and AdaBoost. Additionally, deep learning models, including Neural Networks, were utilised to explore intricate relationships within diabetes-related indicators. Notably, the Gradient Boosting model achieved an impressive accuracy of 99%, with 99% precision, 97% recall and 97% F1-score. To implement these solutions, I used Python as the programming language, employing libraries such as scikit-learn, NumPy, Pandas and Matplotlib, while Streamlit served as the app’s framework.18 0Item Restricted Facial Emotion Recognition via Label Distribution Learning and Customized Convolutional Layers(The University of Warwick, 2024-11) Almowallad, Abeer; Sanchez, VictorThis thesis attempts to investigate the task of recognizing human emotions from facial expressions in images, a topic that has been interest of to researchers in computer vision and machine learning. It addresses the challenge of deciphering a mixture of six basic emotions—happiness, sadness, anger, fear, surprise, and disgust—each presented with distinct intensities. This thesis introduces three Label Distribution Learning (LDL) frameworks to tackle this. Previous studies have dealt with this challenge by using LDL and focusing on optimizing a conditional probability function that attempts to reduce the relative entropy of the predicted distribution with respect to the target distribution, which leads to a lack of generality of the model. First, we propose a deep learning framework for LDL, utilizing convolutional neural network (CNN) features to broaden the model’s generalization capabilities. Named EDL-LBCNN, this framework integrates a Local Binary Convolutional (LBC) layer to refine the texture information extracted from CNNs, targeting a more precise emotion recognition. Secondly, we propose VCNN-ELDL framework, which employs an innovative Visibility Convolutional Layer (VCL). The VCL is engineered to maintain the advantages of traditional convolutional (Conv) layers for feature extraction, while also reducing the number of learnable parameters and enhancing the capture of crucial texture features from facial images. Furthermore, this research presents a novel Transformer architecture, the Visibility Convolutional Vision Transformer (VCLvT), incorporating Depth-Wise Visibility Convolutional Layers (DepthVCL) to bolster spatial feature extraction. This novel approach yields promising outcomes, particularly on limited datasets, showcasing its capacity to meet or exceed state-of-the-art performance across different dataset sizes. Through these advancements, the thesis significantly contributes to the advancement of facial emotion recognition, presenting robust, scalable models adept at interpreting the complex nuances of human emotions.17 0Item Restricted Optimizing Deep Learning Architectures for Enhanced Breast Cancer Detection on Mammography Images(University of Liverpool, 2024) Albalawi, Alaa; Anosova, OlgaBreast cancer is a major health issue affecting millions of women globally, and early detection through mammography is critical for improving survival rates. However, mammography often faces challenges, such as imbalanced datasets and poor image quality, especially in dense breast tissue, which complicates accurate detection. This project explores the use of deep learning techniques, including Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), to address these challenges and enhance breast cancer detection. Five models—ResNet50V2, MobileNetV2, VGG16, ResNet from scratch, and ViT—were compared using various evaluation metrics. Two datasets, RSNA and MIAS, were used, with preprocessing applied only to the RSNA dataset. The experiments were divided into three stages: the first stage evaluated the original RSNA dataset without preprocessing, the second stage tested the balanced and preprocessed RSNA dataset with and without data augmentation, and the third stage applied similar experiments on the MIAS dataset. The results showed that preprocessing and balancing the RSNA dataset significantly improved model performance, while data augmentation further enhanced accuracy and generalization. ViT models outperformed other CNN architectures, demonstrating superior detection abilities after augmentation. ResNet from scratch also showed strong results, benefiting from its controlled architecture that adapted well to high-resolution images. This study highlights how addressing class imbalance and optimising model architectures can lead to more effective breast cancer detection using deep learning.18 0Item Restricted Using Semantic Richness for Metaphor Detection using Deep Learning(University of Birmingham, 2024) Alnafesah, Ghadi; Lee, MarkABSTRACT The Natural Language Processing (NLP) encounters difficulties with metaphors, known for their creative and non-literal usage. Metaphors involve using words or phrases from one context in entirely different contexts, making the meaning less clear and requiring human interpretation for understanding. This dissertation places its focus on the semantic richness elements derived from the perceptual part of the semantic network. These elements serve as the main linguistic features integrated into vector representations. By extracting the semantic information encompassing concreteness, imageability, sensory experience, sentiment, and embodiment, this study seeks to explore the feasibility of detecting metaphors using deep learning models. The investigation is conducted using two experimental structures: sentence-level classification for the categorisation of entire sentences and word-level classification for individual words. These models are assessed across three metaphorical datasets: VUAMC, MOH-X, and TroFi. The main objective is to evaluate the impact of these semantic elements on the metaphor detection task, with the potential for enhancing model performance.35 0Item Restricted Modelling Efficient and Robust Solutions for Microbiology Image Analysis Using Deep Learning(The University of Queensland, 2024-06) Alhammad, Sarah; Lovell, BrianMicroscopic image analysis plays a crucial role in clinical microbiology laboratories for diagnostic purposes. Highly skilled microbiologists, also known as pathologists, are required to interpret vari- ous images, including Gram stain smears. These samples contain vital diagnostic information, such as identifying the presence and types of bacteria, evaluating specimen quality, and cell counting. However, manual interpretation of conventional glass microscopy slides remains a time-consuming, labour-intensive, and operator-dependent process. In high-volume pathology laboratories, implement- ing an artificial intelligent system could offer significant advantages by alleviating limitations faced by conventional pathology on a larger scale. Such a system would ensure enhanced accuracy, reduced workload for pathologists, and improved objectivity and efficiency. Consequently, this has motivated the research using data-driven techniques to develop automated interpretations of pathology images, particularly focusing on Gram stains. With the vast development and advancement in computer vision techniques, researchers have been able to explore the realm of Computer-Aided Diagnoses (CAD). The emergence of deep learning has revolutionised the analysis of pathology and medical images, moving away from traditional handcrafted features to leveraging the power of deep learning algorithms. Among these algorithms, Convolutional Neural Networks (CNNs) have demonstrated their ability to learn features from datasets, leading to enhanced performance and increased robustness of classifiers and detectors against variations Despite the extensive literature on pathology images, the automatic analysis of the Gram stain test using CNNs has not gained the same level of attention as other pathology tests such as breast cancer, lymphoma and colorectal cancer. It is exceedingly rare to find datasets relating to the very important Gram stain, and this data scarcity has likely hindered research on Gram stain automation and limited research in this area. This thesis aims to apply deep learning techniques to analyse pathology images, with a specific focus on Gram stain data. The aim is to discover novel approaches that can enhance the accuracy and efficiency of Gram stain analysis, bridging the gap in research and paving the way for advancements in this critical area. Initially, a CNN-based classifier was proposed for Gram-positive cocci bacteria subtypes in blood cultures. Throughout the study, the effect of downsampling, data augmentation, and image size on classification accuracy and speed was studied. To conduct these experiments, a novel dataset provided by Sullivan Nicolaides Pathology (SNP) consisting of three distinct bacteria subtypes, namely Staphylococcus, Enterococcus and Streptococcus were used. The sub-images were obtained from blood culture WSIs captured by the in-house SNP MicroLab using a ×63 objective without coverslips or oil immersion. The results show that a CNN-based classifier distinguishes between these bacteria subtypes with high classification accuracy. Secondly, existing CNN classification backbones operate under the assumption that all testing classes have been encountered during model training. However, in certain scenarios, it may be infeasible to collect all bacteria subtypes during the model training phase. CNNs are incapable of estimating their uncertainty, and they assume full knowledge of the world. To avoid misdiagnosis risk in the bacteria classification task, OpenGram a framework to open CNN classifier was proposed in this study that aims to tackle the problem of bacteria subtyping from an open-set perspective. Open-set recognition models can classify known instances and detect unknown samples of novel classes. OpenGram combines a CNN classifier with a Gaussian mixtures model to adapt to open-set classification. The results demonstrate OpenGram’s efficacy in accurately detecting unknown bacteria classes that were not encountered by the network during training, while maintaining the ability to classify known bacteria classes. Thirdly, most deep learning-based object detection methods rely on the availability of large sets of annotated training data, assuming that both training and testing data belong to the same feature space. However, these assumptions may not always hold true in real-world applications, particularly in the domain of pathology images. The process of collecting annotations for pathology images can be costly and labor-intensive. Additionally, testing supervised models on different distributions can degrade detector performance as these models might not be properly generalised to other domains. The objective was to tackle this lack of instance-level cell labels in Gram stain WSIs for the epithelial and leukocyte cell counting task. HybridGram, a framework with image translation and pseudo- labelling modules to completely avoid manual labelling on a new dataset was presented. The results demonstrate that HybridGram effectively bridges the performance gap between fully supervised and unsupervised models in this context.26 0